chenxingqiang commited on
Commit
3228ab0
·
1 Parent(s): d6ecb31

Optimize model loading and improve user experience

Browse files
README.md CHANGED
@@ -38,6 +38,36 @@ This Hugging Face Space provides an interactive interface for analyzing radar im
38
  3. View the detection results and analysis report
39
  4. Access previous analyses through the history feature
40
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
41
  ## Technical Details
42
 
43
  - **Model**: PaliGemma-3b fine-tuned for radar defect detection
@@ -52,6 +82,12 @@ The following environment variables need to be set in your Space:
52
  - `HF_TOKEN`: Your Hugging Face token for accessing the model
53
  - `DATABASE_URL` (optional): URL for the database connection
54
 
 
 
 
 
 
 
55
  ## Requirements
56
 
57
  See `requirements.txt` for the complete list of dependencies.
 
38
  3. View the detection results and analysis report
39
  4. Access previous analyses through the history feature
40
 
41
+ ## Setup Instructions
42
+
43
+ ### Local Development
44
+
45
+ 1. Clone this repository:
46
+ ```bash
47
+ git clone https://huggingface.co/spaces/xingqiang/radar-analysis
48
+ cd radar-analysis
49
+ ```
50
+
51
+ 2. Install dependencies:
52
+ ```bash
53
+ pip install -r requirements.txt
54
+ ```
55
+
56
+ 3. Set up environment variables:
57
+ - Create a `.env` file in the root directory
58
+ - Add your Hugging Face token: `HF_TOKEN=your_token_here`
59
+
60
+ 4. Run the application:
61
+ ```bash
62
+ python app.py
63
+ ```
64
+
65
+ ### Hugging Face Space Deployment
66
+
67
+ 1. Fork this repository to your Hugging Face account
68
+ 2. Set the `HF_TOKEN` secret in your Space settings
69
+ 3. Deploy the Space
70
+
71
  ## Technical Details
72
 
73
  - **Model**: PaliGemma-3b fine-tuned for radar defect detection
 
82
  - `HF_TOKEN`: Your Hugging Face token for accessing the model
83
  - `DATABASE_URL` (optional): URL for the database connection
84
 
85
+ ## Troubleshooting
86
+
87
+ - **Memory Issues**: The application will automatically switch to demo mode if there's not enough memory
88
+ - **Model Loading Errors**: Check that your Hugging Face token has access to the required model
89
+ - **Image Processing Errors**: Ensure uploaded images are in a supported format (PNG, JPG)
90
+
91
  ## Requirements
92
 
93
  See `requirements.txt` for the complete list of dependencies.
__pycache__/app.cpython-311.pyc ADDED
Binary file (53 kB). View file
 
__pycache__/config.cpython-311.pyc ADDED
Binary file (762 Bytes). View file
 
__pycache__/feature_extraction.cpython-311.pyc ADDED
Binary file (4.36 kB). View file
 
__pycache__/model.cpython-311.pyc ADDED
Binary file (12.4 kB). View file
 
app.py CHANGED
@@ -12,6 +12,7 @@ import plotly.express as px
12
  import plotly.graph_objects as go
13
  import pandas as pd
14
  from functools import partial
 
15
 
16
  from model import RadarDetectionModel
17
  from feature_extraction import (calculate_amplitude, classify_amplitude,
@@ -22,6 +23,12 @@ from feature_extraction import (calculate_amplitude, classify_amplitude,
22
  from report_generation import generate_report, render_report
23
  from utils import plot_detection
24
  from database import save_report, get_report_history
 
 
 
 
 
 
25
 
26
  # Set theme and styling
27
  THEME = gr.themes.Soft(
@@ -35,32 +42,40 @@ THEME = gr.themes.Soft(
35
  # Create a simple dark mode flag instead of custom theme
36
  DARK_MODE = False
37
 
 
 
 
 
 
 
 
 
38
  class TechnicalReportGenerator:
39
  def __init__(self):
40
  self.timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
41
-
42
  def generate_model_analysis(self, model_outputs):
43
  """Generate model-specific analysis section"""
44
  model_section = "## Model Analysis\n\n"
45
-
46
  # Image encoder analysis
47
  model_section += "### Image Encoder (SigLIP-So400m) Analysis\n"
48
  model_section += "- Feature extraction quality: {:.2f}%\n".format(model_outputs.get('feature_quality', 0) * 100)
49
  model_section += "- Image encoding latency: {:.2f}ms\n".format(model_outputs.get('encoding_latency', 0))
50
  model_section += "- Feature map dimensions: {}\n\n".format(model_outputs.get('feature_dimensions', 'N/A'))
51
-
52
  # Text decoder analysis
53
  model_section += "### Text Decoder (Gemma-2B) Analysis\n"
54
  model_section += "- Text generation confidence: {:.2f}%\n".format(model_outputs.get('text_confidence', 0) * 100)
55
  model_section += "- Decoding latency: {:.2f}ms\n".format(model_outputs.get('decoding_latency', 0))
56
  model_section += "- Token processing rate: {:.2f} tokens/sec\n\n".format(model_outputs.get('token_rate', 0))
57
-
58
  return model_section
59
 
60
  def generate_detection_analysis(self, detection_results):
61
  """Generate detailed detection analysis section"""
62
  detection_section = "## Detection Analysis\n\n"
63
-
64
  # Detection metrics
65
  detection_section += "### Object Detection Metrics\n"
66
  detection_section += "| Metric | Value |\n"
@@ -72,29 +87,29 @@ class TechnicalReportGenerator:
72
  detection_section += "| Processing Time | {:.2f}ms |\n\n".format(
73
  detection_results.get('processing_time', 0)
74
  )
75
-
76
  # Detailed detection results
77
  detection_section += "### Detection Details\n"
78
  detection_section += "| Object | Confidence | Bounding Box |\n"
79
  detection_section += "|--------|------------|---------------|\n"
80
-
81
  boxes = detection_results.get('boxes', [])
82
  scores = detection_results.get('scores', [])
83
  labels = detection_results.get('labels', [])
84
-
85
  for box, score, label in zip(boxes, scores, labels):
86
  detection_section += "| {} | {:.2f}% | {} |\n".format(
87
  label,
88
  score * 100,
89
  [round(coord, 2) for coord in box]
90
  )
91
-
92
  return detection_section
93
 
94
  def generate_multimodal_analysis(self, mm_results):
95
  """Generate multimodal analysis section"""
96
  mm_section = "## Multimodal Analysis\n\n"
97
-
98
  # Feature correlation analysis
99
  mm_section += "### Feature Correlation Analysis\n"
100
  mm_section += "- Text-Image Alignment Score: {:.2f}%\n".format(
@@ -106,19 +121,19 @@ class TechnicalReportGenerator:
106
  mm_section += "- Feature Space Correlation: {:.2f}\n\n".format(
107
  mm_results.get('feature_correlation', 0)
108
  )
109
-
110
  return mm_section
111
 
112
  def generate_performance_metrics(self, perf_data):
113
  """Generate performance metrics section"""
114
  perf_section = "## Performance Metrics\n\n"
115
-
116
  # System metrics
117
  perf_section += "### System Performance\n"
118
  perf_section += "- Total Processing Time: {:.2f}ms\n".format(perf_data.get('total_time', 0))
119
  perf_section += "- Peak Memory Usage: {:.2f}MB\n".format(perf_data.get('peak_memory', 0))
120
  perf_section += "- GPU Utilization: {:.2f}%\n\n".format(perf_data.get('gpu_util', 0))
121
-
122
  # Pipeline metrics
123
  perf_section += "### Pipeline Statistics\n"
124
  perf_section += "| Stage | Time (ms) | Memory (MB) |\n"
@@ -130,108 +145,370 @@ class TechnicalReportGenerator:
130
  stats.get('time', 0),
131
  stats.get('memory', 0)
132
  )
133
-
134
  return perf_section
135
 
136
  def generate_report(self, results):
137
  """Generate comprehensive technical report"""
138
  report = f"# Technical Analysis Report\nGenerated at: {self.timestamp}\n\n"
139
-
140
  # Add model analysis
141
  report += self.generate_model_analysis(results.get('model_outputs', {}))
142
-
143
  # Add detection analysis
144
  report += self.generate_detection_analysis(results.get('detection_results', {}))
145
-
146
  # Add multimodal analysis
147
  report += self.generate_multimodal_analysis(results.get('multimodal_results', {}))
148
-
149
  # Add performance metrics
150
  report += self.generate_performance_metrics(results.get('performance_data', {}))
151
-
152
  return report
153
 
154
- # Initialize model with HF token from environment
155
- model = None
156
- USE_DEMO_MODE = False
157
- MODEL_NAME = "Extremely4606/paligemma24_12_30" # Alternative model instead of Google's gated model
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
158
 
159
- try:
160
- hf_token = os.getenv("HF_TOKEN")
161
- print(f"Attempting to load model: {MODEL_NAME}")
162
- # Try to initialize without token first since this model might be public
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
163
  try:
164
- model = RadarDetectionModel(model_name=MODEL_NAME)
165
- print(f"Successfully loaded model {MODEL_NAME} without authentication")
 
 
166
  except Exception as e:
167
- if not hf_token:
168
- print("Warning: HF_TOKEN environment variable not set. Using demo mode.")
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
169
  USE_DEMO_MODE = True
170
  else:
171
- print(f"Attempting to load model {MODEL_NAME} with authentication")
172
- model = RadarDetectionModel(model_name=MODEL_NAME, use_auth_token=hf_token)
173
- except Exception as e:
174
- print(f"Warning: Model initialization failed: {str(e)}")
175
- print("Falling back to demo mode.")
176
- USE_DEMO_MODE = True
 
 
 
 
177
 
178
  def initialize_model():
179
- global model, USE_DEMO_MODE, MODEL_NAME
180
- if USE_DEMO_MODE:
181
- return None, None # Will use mock data in demo mode
 
182
 
183
- if model is None:
184
- try:
185
- # Try to initialize without token first since this model might be public
186
- try:
187
- model = RadarDetectionModel(model_name=MODEL_NAME)
188
- except Exception as e:
189
- hf_token = os.getenv("HF_TOKEN")
190
- if not hf_token:
191
- USE_DEMO_MODE = True
192
- return None, None
193
-
194
- model = RadarDetectionModel(model_name=MODEL_NAME, use_auth_token=hf_token)
195
- except Exception as e:
 
 
 
 
 
 
 
 
 
 
 
196
  USE_DEMO_MODE = True
197
- return None, None
198
 
199
- return model, None
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
200
 
201
  def create_confidence_chart(scores, labels):
202
  """Create a bar chart for confidence scores"""
203
  if not scores or not labels:
204
  return None
205
-
206
  df = pd.DataFrame({
207
  'Label': labels,
208
  'Confidence': [score * 100 for score in scores]
209
  })
210
-
211
  fig = px.bar(
212
- df,
213
- x='Label',
214
  y='Confidence',
215
  title='Detection Confidence Scores',
216
  labels={'Confidence': 'Confidence (%)'},
217
  color='Confidence',
218
  color_continuous_scale='viridis'
219
  )
220
-
221
  fig.update_layout(
222
  xaxis_title='Detected Object',
223
  yaxis_title='Confidence (%)',
224
  yaxis_range=[0, 100],
225
  template='plotly_white'
226
  )
227
-
228
  return fig
229
 
230
  def create_feature_radar_chart(features):
231
  """Create a radar chart for feature analysis"""
232
  categories = list(features.keys())
233
  values = []
234
-
235
  # Convert text classifications to numeric values (1-5 scale)
236
  for feature in features.values():
237
  if "High" in feature:
@@ -246,16 +523,16 @@ def create_feature_radar_chart(features):
246
  values.append(1)
247
  else:
248
  values.append(0)
249
-
250
  fig = go.Figure()
251
-
252
  fig.add_trace(go.Scatterpolar(
253
  r=values,
254
  theta=categories,
255
  fill='toself',
256
  name='Feature Analysis'
257
  ))
258
-
259
  fig.update_layout(
260
  polar=dict(
261
  radialaxis=dict(
@@ -266,108 +543,109 @@ def create_feature_radar_chart(features):
266
  title='Feature Analysis Radar Chart',
267
  template='plotly_white'
268
  )
269
-
270
  return fig
271
 
272
  def create_heatmap(image_array):
273
  """Create a heatmap visualization of the image intensity"""
274
  if image_array is None:
275
  return None
276
-
277
  # Convert to grayscale if needed
278
  if len(image_array.shape) == 3 and image_array.shape[2] == 3:
279
  gray_img = np.mean(image_array, axis=2)
280
  else:
281
  gray_img = image_array
282
-
283
  fig = px.imshow(
284
  gray_img,
285
  color_continuous_scale='inferno',
286
  title='Signal Intensity Heatmap'
287
  )
288
-
289
  fig.update_layout(
290
  xaxis_title='X Position',
291
  yaxis_title='Y Position',
292
  template='plotly_white'
293
  )
294
-
295
  return fig
296
 
 
 
 
 
 
 
 
 
 
 
 
297
  def process_image_streaming(image, generate_tech_report=False, progress=gr.Progress()):
298
- """Process image with streaming progress updates"""
299
  if image is None:
300
- raise gr.Error("Please upload an image.")
301
 
302
- # Initialize model if needed
303
- progress(0.1, desc="Initializing model...")
 
304
  global model, USE_DEMO_MODE
305
 
306
  if not USE_DEMO_MODE:
307
- model, error = initialize_model()
308
- if error:
309
- progress(0.15, desc="Switching to demo mode...")
310
  USE_DEMO_MODE = True
311
 
312
  try:
313
- # Convert to PIL Image if needed
314
  if isinstance(image, np.ndarray):
315
  image = Image.fromarray(image)
316
 
317
- # Run detection
318
- progress(0.2, desc="Running detection...")
 
319
 
320
  if USE_DEMO_MODE:
321
- # Use mock detection results in demo mode
322
  detection_result = {
323
  'boxes': [[100, 100, 200, 200], [300, 300, 400, 400]],
324
  'scores': [0.92, 0.85],
325
- 'labels': ['Crack', 'Corrosion'],
326
  'image': image
327
  }
328
  else:
329
- detection_result = model.detect(image)
330
-
331
- # Extract features
332
- progress(0.3, desc="Extracting features...")
333
- np_image = np.array(image)
334
- amplitude = calculate_amplitude(np_image)
335
- amplitude_class = classify_amplitude(amplitude)
336
-
337
- if len(detection_result['boxes']) > 0:
338
- box = detection_result['boxes'][0]
339
- if not isinstance(box, list):
340
- box = box.tolist()
341
- distribution_range = calculate_distribution_range(box)
342
- distribution_class = classify_distribution_range(distribution_range)
343
- else:
344
- distribution_class = "No defects detected"
345
-
346
- attenuation_rate = calculate_attenuation_rate(np_image)
347
- attenuation_class = classify_attenuation_rate(attenuation_rate)
348
-
349
- reflection_count = count_reflections(np_image)
350
- reflection_class = classify_reflections(reflection_count)
351
 
352
- features = {
353
- "Amplitude": amplitude_class,
354
- "Distribution Range": distribution_class,
355
- "Attenuation Rate": attenuation_class,
356
- "Reflection Count": reflection_class
357
- }
358
 
359
- # Create visualization charts
360
- progress(0.5, desc="Creating visualizations...")
361
  confidence_chart = create_confidence_chart(
362
- detection_result.get('scores', []),
363
  detection_result.get('labels', [])
364
  )
365
 
366
  feature_chart = create_feature_radar_chart(features)
367
- heatmap = create_heatmap(np_image)
368
 
369
- # Start performance tracking
370
- progress(0.6, desc="Analyzing performance...")
371
  start_time = time.time()
372
  performance_data = {
373
  'pipeline_stats': {},
@@ -375,7 +653,7 @@ def process_image_streaming(image, generate_tech_report=False, progress=gr.Progr
375
  'gpu_util': 0
376
  }
377
 
378
- # Process image and get results
379
  stage_start = time.time()
380
  detection_results = detection_result
381
  detection_results['processing_time'] = (time.time() - stage_start) * 1000
@@ -384,7 +662,7 @@ def process_image_streaming(image, generate_tech_report=False, progress=gr.Progr
384
  'memory': get_memory_usage()
385
  }
386
 
387
- # Extract features and analyze
388
  stage_start = time.time()
389
  model_outputs = {
390
  'feature_quality': 0.85,
@@ -399,7 +677,7 @@ def process_image_streaming(image, generate_tech_report=False, progress=gr.Progr
399
  'memory': get_memory_usage()
400
  }
401
 
402
- # Perform multimodal analysis
403
  stage_start = time.time()
404
  multimodal_results = {
405
  'alignment_score': 0.78,
@@ -411,20 +689,20 @@ def process_image_streaming(image, generate_tech_report=False, progress=gr.Progr
411
  'memory': get_memory_usage()
412
  }
413
 
414
- # Update performance data
415
  performance_data['total_time'] = (time.time() - start_time) * 1000
416
  performance_data['peak_memory'] = get_peak_memory_usage()
417
  performance_data['gpu_util'] = get_gpu_utilization()
418
 
419
- # Generate analysis report
420
- progress(0.8, desc="Generating reports...")
421
  analysis_report = generate_report(detection_result, features)
422
 
423
- # Prepare output
424
  output_image = plot_detection(image, detection_result)
425
 
426
  if generate_tech_report:
427
- # Prepare results for technical report
428
  tech_report_data = {
429
  'model_outputs': model_outputs,
430
  'detection_results': detection_results,
@@ -432,23 +710,29 @@ def process_image_streaming(image, generate_tech_report=False, progress=gr.Progr
432
  'performance_data': performance_data
433
  }
434
 
435
- # Generate technical report
436
  tech_report = TechnicalReportGenerator().generate_report(tech_report_data)
437
 
438
- # Save technical report to temporary file
439
  report_path = "temp_tech_report.md"
440
  with open(report_path, "w") as f:
441
  f.write(tech_report)
442
 
443
- progress(1.0, desc="Analysis complete!")
 
 
444
  return output_image, analysis_report, report_path, confidence_chart, feature_chart, heatmap
445
 
446
- progress(1.0, desc="Analysis complete!")
 
 
447
  return output_image, analysis_report, None, confidence_chart, feature_chart, heatmap
448
 
449
  except Exception as e:
450
- error_msg = f"Error processing image: {str(e)}"
451
  print(error_msg)
 
 
452
  raise gr.Error(error_msg)
453
 
454
  def display_history():
@@ -472,12 +756,25 @@ def display_history():
472
  def get_memory_usage():
473
  """Get current memory usage in MB"""
474
  process = psutil.Process()
475
- return process.memory_info().rss / 1024 / 1024
 
476
 
477
  def get_peak_memory_usage():
478
  """Get peak memory usage in MB"""
479
- process = psutil.Process()
480
- return process.memory_info().peak_wset / 1024 / 1024 if hasattr(process.memory_info(), 'peak_wset') else 0
 
 
 
 
 
 
 
 
 
 
 
 
481
 
482
  def get_gpu_utilization():
483
  """Get GPU utilization percentage"""
@@ -488,128 +785,274 @@ def get_gpu_utilization():
488
  pass
489
  return 0
490
 
 
 
 
 
 
 
 
491
  def toggle_dark_mode():
492
  """Toggle between light and dark themes"""
493
  global DARK_MODE
494
  DARK_MODE = not DARK_MODE
495
  return gr.Theme.darkmode() if DARK_MODE else THEME
496
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
497
  # Create Gradio interface
498
  with gr.Blocks(theme=THEME) as iface:
499
  theme_state = gr.State(THEME)
500
-
501
  with gr.Row():
502
- gr.Markdown("# Radar Image Analysis System")
503
- dark_mode_btn = gr.Button("🌓 Toggle Dark Mode", scale=0)
504
-
 
 
 
 
 
 
 
 
505
  if USE_DEMO_MODE:
506
- gr.Markdown("""
507
- ### ⚠️ Running in Demo Mode
508
- The app is currently running in demo mode because access to the PaliGemma model is restricted.
509
-
510
- To use the full model capabilities:
511
- 1. Get access to the PaliGemma model at [Hugging Face](https://huggingface.co/google/paligemma-3b-ft-coco35l-224)
512
- 2. Add your Hugging Face token as an environment variable named `HF_TOKEN` in the Space settings
513
-
514
- Demo mode will show simulated results for demonstration purposes.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
515
  """, elem_id="demo-mode-warning")
516
-
517
- gr.Markdown("Upload a radar image to analyze defects and generate technical reports")
518
-
519
  with gr.Tabs() as tabs:
520
- with gr.TabItem("Analysis", id="analysis"):
521
  with gr.Row():
522
  with gr.Column(scale=1):
523
- with gr.Accordion("Input", open=True):
524
  input_image = gr.Image(
525
- type="pil",
526
- label="Upload Radar Image",
527
  elem_id="input-image",
528
  sources=["upload", "webcam", "clipboard"],
529
  tool="editor"
530
  )
531
  tech_report_checkbox = gr.Checkbox(
532
- label="Generate Technical Report",
533
  value=False,
534
- info="Creates a detailed technical analysis report"
535
  )
536
  analyze_button = gr.Button(
537
- "Analyze",
538
  variant="primary",
539
  elem_id="analyze-btn"
540
  )
541
-
542
  with gr.Column(scale=2):
543
- with gr.Accordion("Detection Results", open=True):
544
  output_image = gr.Image(
545
- type="pil",
546
- label="Detection Result",
547
  elem_id="output-image"
548
  )
549
-
550
- with gr.Accordion("Analysis Report", open=True):
551
  output_report = gr.HTML(
552
- label="Analysis Report",
553
  elem_id="analysis-report"
554
  )
555
  tech_report_output = gr.File(
556
- label="Technical Report",
557
  elem_id="tech-report"
558
  )
559
-
560
  with gr.Row():
561
  with gr.Column():
562
  confidence_plot = gr.Plot(
563
- label="Confidence Scores",
564
  elem_id="confidence-plot"
565
  )
566
-
567
  with gr.Column():
568
  feature_plot = gr.Plot(
569
- label="Feature Analysis",
570
  elem_id="feature-plot"
571
  )
572
-
573
  with gr.Row():
574
  heatmap_plot = gr.Plot(
575
- label="Signal Intensity Heatmap",
576
  elem_id="heatmap-plot"
577
  )
578
-
579
- with gr.TabItem("History", id="history"):
580
  with gr.Row():
581
- history_button = gr.Button("Refresh History")
582
  history_output = gr.HTML(elem_id="history-output")
583
-
584
- with gr.TabItem("Help", id="help"):
585
  gr.Markdown("""
586
- ## How to Use This Tool
587
-
588
- 1. **Upload an Image**: Click the upload button to select a radar image for analysis
589
- 2. **Generate Technical Report** (Optional): Check this box if you want a detailed technical report
590
- 3. **Analyze**: Click the Analyze button to process the image
591
- 4. **View Results**:
592
- - The detection visualization shows identified defects
593
- - The analysis report provides a summary of findings
594
- - The technical report (if requested) offers detailed metrics
595
- - Charts provide visual representation of confidence scores and feature analysis
596
-
597
- ## About the Model
598
-
599
- This system uses PaliGemma, a vision-language model that combines SigLIP-So400m (image encoder) and Gemma-2B (text decoder) for joint object detection and multimodal analysis.
600
-
601
- ## Keyboard Shortcuts
602
-
603
- - **Ctrl+A**: Trigger analysis
604
- - **Ctrl+D**: Toggle dark mode
605
-
606
- ## Troubleshooting
607
-
608
- - If the analysis fails, try uploading a different image format
609
- - Ensure the image is a valid radar scan
610
- - For technical issues, check the console logs
611
  """)
612
-
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
613
  # Set up event handlers
614
  dark_mode_btn.click(
615
  fn=toggle_dark_mode,
@@ -617,21 +1060,21 @@ with gr.Blocks(theme=THEME) as iface:
617
  outputs=[iface],
618
  api_name="toggle_theme"
619
  )
620
-
621
  analyze_button.click(
622
  fn=process_image_streaming,
623
  inputs=[input_image, tech_report_checkbox],
624
  outputs=[output_image, output_report, tech_report_output, confidence_plot, feature_plot, heatmap_plot],
625
  api_name="analyze"
626
  )
627
-
628
  history_button.click(
629
  fn=display_history,
630
  inputs=[],
631
  outputs=[history_output],
632
  api_name="history"
633
  )
634
-
635
  # Add keyboard shortcuts
636
  iface.load(lambda: None, None, None, _js="""
637
  () => {
@@ -640,11 +1083,11 @@ with gr.Blocks(theme=THEME) as iface:
640
  document.getElementById('analyze-btn').click();
641
  }
642
  if (e.key === 'd' && e.ctrlKey) {
643
- document.querySelector('button:contains("Toggle Dark Mode")').click();
644
  }
645
  });
646
  }
647
  """)
648
 
649
  # Launch the interface
650
- iface.launch()
 
12
  import plotly.graph_objects as go
13
  import pandas as pd
14
  from functools import partial
15
+ import logging
16
 
17
  from model import RadarDetectionModel
18
  from feature_extraction import (calculate_amplitude, classify_amplitude,
 
23
  from report_generation import generate_report, render_report
24
  from utils import plot_detection
25
  from database import save_report, get_report_history
26
+ from config import MODEL_NAME
27
+
28
+ # Configure logging
29
+ logging.basicConfig(level=logging.INFO,
30
+ format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
31
+ logger = logging.getLogger(__name__)
32
 
33
  # Set theme and styling
34
  THEME = gr.themes.Soft(
 
42
  # Create a simple dark mode flag instead of custom theme
43
  DARK_MODE = False
44
 
45
+ # Global variables
46
+ model = None
47
+ USE_DEMO_MODE = False
48
+ HF_TOKEN = os.environ.get("HF_TOKEN") or os.environ.get("HF_TOCKEN")
49
+
50
+ # 添加一个标志,表示是否已经尝试过初始化模型
51
+ MODEL_INIT_ATTEMPTED = False
52
+
53
  class TechnicalReportGenerator:
54
  def __init__(self):
55
  self.timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
56
+
57
  def generate_model_analysis(self, model_outputs):
58
  """Generate model-specific analysis section"""
59
  model_section = "## Model Analysis\n\n"
60
+
61
  # Image encoder analysis
62
  model_section += "### Image Encoder (SigLIP-So400m) Analysis\n"
63
  model_section += "- Feature extraction quality: {:.2f}%\n".format(model_outputs.get('feature_quality', 0) * 100)
64
  model_section += "- Image encoding latency: {:.2f}ms\n".format(model_outputs.get('encoding_latency', 0))
65
  model_section += "- Feature map dimensions: {}\n\n".format(model_outputs.get('feature_dimensions', 'N/A'))
66
+
67
  # Text decoder analysis
68
  model_section += "### Text Decoder (Gemma-2B) Analysis\n"
69
  model_section += "- Text generation confidence: {:.2f}%\n".format(model_outputs.get('text_confidence', 0) * 100)
70
  model_section += "- Decoding latency: {:.2f}ms\n".format(model_outputs.get('decoding_latency', 0))
71
  model_section += "- Token processing rate: {:.2f} tokens/sec\n\n".format(model_outputs.get('token_rate', 0))
72
+
73
  return model_section
74
 
75
  def generate_detection_analysis(self, detection_results):
76
  """Generate detailed detection analysis section"""
77
  detection_section = "## Detection Analysis\n\n"
78
+
79
  # Detection metrics
80
  detection_section += "### Object Detection Metrics\n"
81
  detection_section += "| Metric | Value |\n"
 
87
  detection_section += "| Processing Time | {:.2f}ms |\n\n".format(
88
  detection_results.get('processing_time', 0)
89
  )
90
+
91
  # Detailed detection results
92
  detection_section += "### Detection Details\n"
93
  detection_section += "| Object | Confidence | Bounding Box |\n"
94
  detection_section += "|--------|------------|---------------|\n"
95
+
96
  boxes = detection_results.get('boxes', [])
97
  scores = detection_results.get('scores', [])
98
  labels = detection_results.get('labels', [])
99
+
100
  for box, score, label in zip(boxes, scores, labels):
101
  detection_section += "| {} | {:.2f}% | {} |\n".format(
102
  label,
103
  score * 100,
104
  [round(coord, 2) for coord in box]
105
  )
106
+
107
  return detection_section
108
 
109
  def generate_multimodal_analysis(self, mm_results):
110
  """Generate multimodal analysis section"""
111
  mm_section = "## Multimodal Analysis\n\n"
112
+
113
  # Feature correlation analysis
114
  mm_section += "### Feature Correlation Analysis\n"
115
  mm_section += "- Text-Image Alignment Score: {:.2f}%\n".format(
 
121
  mm_section += "- Feature Space Correlation: {:.2f}\n\n".format(
122
  mm_results.get('feature_correlation', 0)
123
  )
124
+
125
  return mm_section
126
 
127
  def generate_performance_metrics(self, perf_data):
128
  """Generate performance metrics section"""
129
  perf_section = "## Performance Metrics\n\n"
130
+
131
  # System metrics
132
  perf_section += "### System Performance\n"
133
  perf_section += "- Total Processing Time: {:.2f}ms\n".format(perf_data.get('total_time', 0))
134
  perf_section += "- Peak Memory Usage: {:.2f}MB\n".format(perf_data.get('peak_memory', 0))
135
  perf_section += "- GPU Utilization: {:.2f}%\n\n".format(perf_data.get('gpu_util', 0))
136
+
137
  # Pipeline metrics
138
  perf_section += "### Pipeline Statistics\n"
139
  perf_section += "| Stage | Time (ms) | Memory (MB) |\n"
 
145
  stats.get('time', 0),
146
  stats.get('memory', 0)
147
  )
148
+
149
  return perf_section
150
 
151
  def generate_report(self, results):
152
  """Generate comprehensive technical report"""
153
  report = f"# Technical Analysis Report\nGenerated at: {self.timestamp}\n\n"
154
+
155
  # Add model analysis
156
  report += self.generate_model_analysis(results.get('model_outputs', {}))
157
+
158
  # Add detection analysis
159
  report += self.generate_detection_analysis(results.get('detection_results', {}))
160
+
161
  # Add multimodal analysis
162
  report += self.generate_multimodal_analysis(results.get('multimodal_results', {}))
163
+
164
  # Add performance metrics
165
  report += self.generate_performance_metrics(results.get('performance_data', {}))
166
+
167
  return report
168
 
169
+ def check_available_memory():
170
+ """Check available system memory in MB"""
171
+ try:
172
+ import psutil
173
+ vm = psutil.virtual_memory()
174
+ available_mb = vm.available / (1024 * 1024)
175
+ total_mb = vm.total / (1024 * 1024)
176
+ print(f"Available memory: {available_mb:.2f}MB out of {total_mb:.2f}MB total")
177
+ return available_mb
178
+ except Exception as e:
179
+ print(f"Error checking memory: {str(e)}")
180
+ return 0
181
+
182
+ def monitor_memory_during_loading(model_name, use_auth_token=None):
183
+ """Monitor memory usage during model loading and abort if it gets too high"""
184
+ global USE_DEMO_MODE
185
+
186
+ try:
187
+ # Initial memory check
188
+ initial_memory = get_memory_usage()
189
+ print(f"Initial memory usage: {initial_memory:.2f}MB")
190
+
191
+ # Start loading processor
192
+ print(f"Loading processor from {model_name}")
193
+ if use_auth_token:
194
+ processor = AutoProcessor.from_pretrained(model_name, use_auth_token=use_auth_token)
195
+ else:
196
+ processor = AutoProcessor.from_pretrained(model_name)
197
+
198
+ # Check memory after processor loading
199
+ after_processor_memory = get_memory_usage()
200
+ print(f"Memory after processor loading: {after_processor_memory:.2f}MB (Δ: {after_processor_memory - initial_memory:.2f}MB)")
201
+
202
+ # Check if memory is getting too high
203
+ available_memory = check_available_memory()
204
+ if available_memory < 4000: # Less than 4GB available
205
+ print(f"Warning: Only {available_memory:.2f}MB memory available after loading processor")
206
+ print("Aborting model loading to avoid out-of-memory error")
207
+ USE_DEMO_MODE = True
208
+ return None, None
209
+
210
+ # Start loading model with 8-bit quantization
211
+ print(f"Loading model from {model_name} with 8-bit quantization")
212
+ if use_auth_token:
213
+ model = AutoModelForVision2Seq.from_pretrained(
214
+ model_name,
215
+ use_auth_token=use_auth_token,
216
+ load_in_8bit=True,
217
+ device_map="auto"
218
+ )
219
+ else:
220
+ model = AutoModelForVision2Seq.from_pretrained(
221
+ model_name,
222
+ load_in_8bit=True,
223
+ device_map="auto"
224
+ )
225
 
226
+ # Check memory after model loading
227
+ after_model_memory = get_memory_usage()
228
+ print(f"Memory after model loading: {after_model_memory:.2f}MB (Δ: {after_model_memory - after_processor_memory:.2f}MB)")
229
+
230
+ # Set model to evaluation mode
231
+ model.eval()
232
+
233
+ return processor, model
234
+ except Exception as e:
235
+ print(f"Error during monitored model loading: {str(e)}")
236
+ USE_DEMO_MODE = True
237
+ return None, None
238
+
239
+ def is_running_in_space():
240
+ """Check if we're running in a Hugging Face Space environment"""
241
+ return os.environ.get("SPACE_ID") is not None
242
+
243
+ def is_container_environment():
244
+ """Check if we're running in a container environment"""
245
+ return os.path.exists("/.dockerenv") or os.path.exists("/run/.containerenv")
246
+
247
+ def is_cpu_only():
248
+ """Check if we're running in a CPU-only environment"""
249
+ return not torch.cuda.is_available()
250
+
251
+ def is_low_memory_environment():
252
+ """Check if we're running in a low-memory environment"""
253
+ available_memory = check_available_memory()
254
+ return available_memory < 8000 # Less than 8GB available
255
+
256
+ def is_development_environment():
257
+ """Check if we're running in a development environment"""
258
+ return not (is_running_in_space() or is_container_environment())
259
+
260
+ def is_debug_mode():
261
+ """Check if we're running in debug mode"""
262
+ return os.environ.get("DEBUG", "").lower() in ("1", "true", "yes")
263
+
264
+ def is_test_mode():
265
+ """Check if we're running in test mode"""
266
+ return os.environ.get("TEST", "").lower() in ("1", "true", "yes")
267
+
268
+ def is_low_memory_container():
269
+ """Check if we're running in a container with memory limits"""
270
+ if not is_container_environment():
271
+ return False
272
+
273
+ # Check if cgroup memory limit is set
274
+ try:
275
+ with open('/sys/fs/cgroup/memory/memory.limit_in_bytes', 'r') as f:
276
+ limit = int(f.read().strip())
277
+ # Convert to MB
278
+ limit_mb = limit / (1024 * 1024)
279
+ print(f"Container memory limit: {limit_mb:.2f}MB")
280
+ return limit_mb < 20000 # Less than 20GB
281
+ except:
282
+ # If we can't read the limit, assume it's a low-memory container
283
+ return True
284
+
285
+ def is_space_hardware_type(hardware_type):
286
+ """Check if we're running in a Hugging Face Space with a specific hardware type"""
287
+ if not is_running_in_space():
288
+ return False
289
+
290
+ # Check if SPACE_HARDWARE environment variable matches the specified type
291
+ return os.environ.get("SPACE_HARDWARE", "").lower() == hardware_type.lower()
292
+
293
+ def get_space_hardware_tier():
294
+ """Get the hardware tier of the Hugging Face Space"""
295
+ if not is_running_in_space():
296
+ return "Not a Space"
297
+
298
+ hardware = os.environ.get("SPACE_HARDWARE", "unknown")
299
+
300
+ # Determine the tier based on hardware type
301
+ if hardware.lower() == "cpu":
302
+ return "Basic (CPU)"
303
+ elif hardware.lower() == "t4-small":
304
+ return "Basic (GPU)"
305
+ elif hardware.lower() == "t4-medium":
306
+ return "Standard"
307
+ elif hardware.lower() == "a10g-small":
308
+ return "Pro"
309
+ elif hardware.lower() == "a10g-large":
310
+ return "Pro+"
311
+ elif hardware.lower() == "a100-large":
312
+ return "Enterprise"
313
+ else:
314
+ return f"Unknown ({hardware})"
315
+
316
+ def get_space_hardware_memory():
317
+ """Get the memory size of the Hugging Face Space hardware in GB"""
318
+ if not is_running_in_space():
319
+ return 0
320
+
321
+ hardware = os.environ.get("SPACE_HARDWARE", "unknown").lower()
322
+
323
+ # Determine the memory size based on hardware type
324
+ if hardware == "cpu":
325
+ return 16 # 16GB for CPU
326
+ elif hardware == "t4-small":
327
+ return 16 # 16GB for T4 Small
328
+ elif hardware == "t4-medium":
329
+ return 16 # 16GB for T4 Medium
330
+ elif hardware == "a10g-small":
331
+ return 24 # 24GB for A10G Small
332
+ elif hardware == "a10g-large":
333
+ return 40 # 40GB for A10G Large
334
+ elif hardware == "a100-large":
335
+ return 80 # 80GB for A100 Large
336
+ else:
337
+ return 16 # Default to 16GB
338
+
339
+ def get_total_system_memory():
340
+ """Get total system memory in MB"""
341
  try:
342
+ import psutil
343
+ total_bytes = psutil.virtual_memory().total
344
+ total_mb = total_bytes / (1024 * 1024)
345
+ return total_mb
346
  except Exception as e:
347
+ print(f"Error getting total system memory: {str(e)}")
348
+ return 0
349
+
350
+ def estimate_model_memory_requirements():
351
+ """Estimate the memory requirements for the model"""
352
+ # This is a placeholder implementation. You might want to implement a more accurate estimation based on your model's architecture and typical input sizes.
353
+ try:
354
+ HF_TOCKEN = os.getenv("HF_TOCKEN")
355
+
356
+ # Print startup message
357
+ print("===== Application Startup at", datetime.now().strftime("%Y-%m-%d %H:%M:%S"), "=====")
358
+
359
+ # Get system memory information
360
+ total_memory = get_total_system_memory()
361
+ required_memory = estimate_model_memory_requirements()
362
+ recommended_tier = get_recommended_space_tier()
363
+ print(f"NOTICE: Total system memory: {total_memory:.2f}MB")
364
+ print(f"NOTICE: Estimated model memory requirement: {required_memory:.2f}MB")
365
+ print(f"NOTICE: Recommended Space tier: {recommended_tier}")
366
+
367
+ if is_test_mode():
368
+ print("NOTICE: Running in TEST mode")
369
+ print("NOTICE: Using mock data and responses")
370
+ USE_DEMO_MODE = True
371
+
372
+ if is_debug_mode():
373
+ print("NOTICE: Running in DEBUG mode")
374
+ print("NOTICE: Additional logging and diagnostics will be enabled")
375
+
376
+ if is_development_environment():
377
+ print("NOTICE: Running in development environment")
378
+ print("NOTICE: Full model capabilities may be available depending on system resources")
379
+
380
+ if is_running_in_space():
381
+ print("NOTICE: Running in Hugging Face Space environment")
382
+
383
+ # Check Space hardware type
384
+ hardware_type = get_space_hardware_type()
385
+ hardware_tier = get_space_hardware_tier()
386
+ hardware_memory = get_space_hardware_memory()
387
+ print(f"NOTICE: Space hardware type: {hardware_type} (Tier: {hardware_tier}, Memory: {hardware_memory}GB)")
388
+
389
+ if has_enough_memory_for_model():
390
+ print("NOTICE: This Space has enough memory for the model, but we're still forcing demo mode for stability")
391
+ else:
392
+ print(f"NOTICE: This Space does NOT have enough memory for the model (Need: {required_memory:.2f}MB, Have: {hardware_memory*1024:.2f}MB)")
393
+ print(f"NOTICE: Recommended Space tier: {recommended_tier}")
394
+
395
+ print("NOTICE: FORCING DEMO MODE to avoid 'Memory limit exceeded (16Gi)' error")
396
+ print("NOTICE: The PaliGemma model is too large for the 16GB memory limit in Spaces")
397
+ print("NOTICE: To use the full model, please run this application locally")
398
+ USE_DEMO_MODE = True
399
+ elif is_container_environment():
400
+ print("NOTICE: Running in a container environment")
401
+ print("NOTICE: Memory limits may be enforced by the container runtime")
402
+
403
+ if is_cpu_only():
404
+ print("NOTICE: Running in CPU-only environment")
405
+ print("NOTICE: Model loading and inference will be slower")
406
+
407
+ # Check available memory
408
+ available_memory = check_available_memory()
409
+ print(f"NOTICE: Available memory: {available_memory:.2f}MB")
410
+
411
+ if is_low_memory_environment() and not USE_DEMO_MODE:
412
+ print("NOTICE: Running in a low-memory environment")
413
+ print("NOTICE: Enabling DEMO MODE to avoid memory issues")
414
  USE_DEMO_MODE = True
415
  else:
416
+ # Check available memory before loading
417
+ available_memory = check_available_memory()
418
+ if available_memory < 8000: # If less than 8GB available
419
+ print(f"Warning: Only {available_memory:.2f}MB memory available, which may not be enough for the full model")
420
+ return required_memory
421
+ except Exception as e:
422
+ print(f"Warning: Model initialization failed: {str(e)}")
423
+ print("Falling back to demo mode.")
424
+ USE_DEMO_MODE = True
425
+ return 0
426
 
427
  def initialize_model():
428
+ """
429
+ 仅在需要时初始化模型,不会在应用启动时自动加载
430
+ """
431
+ global model, USE_DEMO_MODE, MODEL_INIT_ATTEMPTED
432
 
433
+ # 如果已经初始化过模型,直接返回
434
+ if model is not None:
435
+ return model
436
+
437
+ # 如果已经尝试过初始化并失败,使用演示模式
438
+ if MODEL_INIT_ATTEMPTED and model is None:
439
+ logger.info("已尝试过初始化模型但失败,使用演示模式")
440
+ USE_DEMO_MODE = True
441
+ return None
442
+
443
+ # 标记为已尝试初始化
444
+ MODEL_INIT_ATTEMPTED = True
445
+
446
+ # 检查是否在Hugging Face Space环境中运行
447
+ if is_running_in_space():
448
+ logger.info("在Hugging Face Space环境中运行")
449
+
450
+ # 检查可用内存
451
+ available_memory = check_available_memory()
452
+ logger.info(f"可用内存: {available_memory:.2f}MB")
453
+
454
+ if available_memory < 8000: # 如果可用内存少于8GB
455
+ logger.warning(f"只有{available_memory:.2f}MB可用内存,可能不足以加载模型")
456
+ logger.info("使用演示模式以避免内存问题")
457
  USE_DEMO_MODE = True
458
+ return None
459
 
460
+ if USE_DEMO_MODE:
461
+ logger.info("使用演示模式 - 不会加载模型")
462
+ return None # 在演示模式下使用模拟数据
463
+
464
+ try:
465
+ # 从环境变量获取token
466
+ hf_token = os.environ.get("HF_TOKEN") or os.environ.get("HF_TOCKEN")
467
+
468
+ logger.info(f"尝试加载模型 {MODEL_NAME}")
469
+ model = RadarDetectionModel(model_name=MODEL_NAME, use_auth_token=hf_token)
470
+ logger.info(f"成功加载模型 {MODEL_NAME}")
471
+ return model
472
+ except Exception as e:
473
+ logger.error(f"模型初始化错误: {str(e)}")
474
+ logger.info("由于模型加载错误,切换到演示模式")
475
+ USE_DEMO_MODE = True
476
+ return None
477
 
478
  def create_confidence_chart(scores, labels):
479
  """Create a bar chart for confidence scores"""
480
  if not scores or not labels:
481
  return None
482
+
483
  df = pd.DataFrame({
484
  'Label': labels,
485
  'Confidence': [score * 100 for score in scores]
486
  })
487
+
488
  fig = px.bar(
489
+ df,
490
+ x='Label',
491
  y='Confidence',
492
  title='Detection Confidence Scores',
493
  labels={'Confidence': 'Confidence (%)'},
494
  color='Confidence',
495
  color_continuous_scale='viridis'
496
  )
497
+
498
  fig.update_layout(
499
  xaxis_title='Detected Object',
500
  yaxis_title='Confidence (%)',
501
  yaxis_range=[0, 100],
502
  template='plotly_white'
503
  )
504
+
505
  return fig
506
 
507
  def create_feature_radar_chart(features):
508
  """Create a radar chart for feature analysis"""
509
  categories = list(features.keys())
510
  values = []
511
+
512
  # Convert text classifications to numeric values (1-5 scale)
513
  for feature in features.values():
514
  if "High" in feature:
 
523
  values.append(1)
524
  else:
525
  values.append(0)
526
+
527
  fig = go.Figure()
528
+
529
  fig.add_trace(go.Scatterpolar(
530
  r=values,
531
  theta=categories,
532
  fill='toself',
533
  name='Feature Analysis'
534
  ))
535
+
536
  fig.update_layout(
537
  polar=dict(
538
  radialaxis=dict(
 
543
  title='Feature Analysis Radar Chart',
544
  template='plotly_white'
545
  )
546
+
547
  return fig
548
 
549
  def create_heatmap(image_array):
550
  """Create a heatmap visualization of the image intensity"""
551
  if image_array is None:
552
  return None
553
+
554
  # Convert to grayscale if needed
555
  if len(image_array.shape) == 3 and image_array.shape[2] == 3:
556
  gray_img = np.mean(image_array, axis=2)
557
  else:
558
  gray_img = image_array
559
+
560
  fig = px.imshow(
561
  gray_img,
562
  color_continuous_scale='inferno',
563
  title='Signal Intensity Heatmap'
564
  )
565
+
566
  fig.update_layout(
567
  xaxis_title='X Position',
568
  yaxis_title='Y Position',
569
  template='plotly_white'
570
  )
571
+
572
  return fig
573
 
574
+ def cleanup_memory():
575
+ """Attempt to clean up memory by forcing garbage collection"""
576
+ try:
577
+ import gc
578
+ gc.collect()
579
+ if torch.cuda.is_available():
580
+ torch.cuda.empty_cache()
581
+ print("Memory cleanup performed")
582
+ except Exception as e:
583
+ print(f"Error during memory cleanup: {str(e)}")
584
+
585
  def process_image_streaming(image, generate_tech_report=False, progress=gr.Progress()):
586
+ """处理图像并提供流式进度更新"""
587
  if image is None:
588
+ raise gr.Error("请上传一张图像。")
589
 
590
+ # 仅在需要时初始化模型
591
+ progress(0.1, desc="初始化模型...")
592
+ log_memory_usage("在process_image中初始化模型之前")
593
  global model, USE_DEMO_MODE
594
 
595
  if not USE_DEMO_MODE:
596
+ model = initialize_model()
597
+ if model is None:
598
+ progress(0.15, desc="切换到演示模式...")
599
  USE_DEMO_MODE = True
600
 
601
  try:
602
+ # 如果需要,将图像转换为PIL Image
603
  if isinstance(image, np.ndarray):
604
  image = Image.fromarray(image)
605
 
606
+ # 运行检测
607
+ progress(0.2, desc="运行检测...")
608
+ log_memory_usage("检测之前")
609
 
610
  if USE_DEMO_MODE:
611
+ # 在演示模式下使用模拟检测结果
612
  detection_result = {
613
  'boxes': [[100, 100, 200, 200], [300, 300, 400, 400]],
614
  'scores': [0.92, 0.85],
615
+ 'labels': ['裂缝', '腐蚀'],
616
  'image': image
617
  }
618
  else:
619
+ try:
620
+ detection_result = model.detect(image)
621
+ log_memory_usage("检测之后")
622
+ except Exception as e:
623
+ logger.error(f"检测过程中出错: {str(e)}")
624
+ # 如果检测失败,切换到演示模式
625
+ USE_DEMO_MODE = True
626
+ detection_result = {
627
+ 'boxes': [[100, 100, 200, 200], [300, 300, 400, 400]],
628
+ 'scores': [0.92, 0.85],
629
+ 'labels': ['错误', '备用'],
630
+ 'image': image
631
+ }
 
 
 
 
 
 
 
 
 
632
 
633
+ # 提取特征
634
+ progress(0.3, desc="提取特征...")
635
+ features = extract_features(image, detection_result)
 
 
 
636
 
637
+ # 创建可视化图表
638
+ progress(0.5, desc="创建可视化...")
639
  confidence_chart = create_confidence_chart(
640
+ detection_result.get('scores', []),
641
  detection_result.get('labels', [])
642
  )
643
 
644
  feature_chart = create_feature_radar_chart(features)
645
+ heatmap = create_heatmap(np.array(image))
646
 
647
+ # 开始性能跟踪
648
+ progress(0.6, desc="分析性能...")
649
  start_time = time.time()
650
  performance_data = {
651
  'pipeline_stats': {},
 
653
  'gpu_util': 0
654
  }
655
 
656
+ # 处理图像并获取结果
657
  stage_start = time.time()
658
  detection_results = detection_result
659
  detection_results['processing_time'] = (time.time() - stage_start) * 1000
 
662
  'memory': get_memory_usage()
663
  }
664
 
665
+ # 提取特征并分析
666
  stage_start = time.time()
667
  model_outputs = {
668
  'feature_quality': 0.85,
 
677
  'memory': get_memory_usage()
678
  }
679
 
680
+ # 执行多模态分析
681
  stage_start = time.time()
682
  multimodal_results = {
683
  'alignment_score': 0.78,
 
689
  'memory': get_memory_usage()
690
  }
691
 
692
+ # 更新性能数据
693
  performance_data['total_time'] = (time.time() - start_time) * 1000
694
  performance_data['peak_memory'] = get_peak_memory_usage()
695
  performance_data['gpu_util'] = get_gpu_utilization()
696
 
697
+ # 生成分析报告
698
+ progress(0.8, desc="生成报告...")
699
  analysis_report = generate_report(detection_result, features)
700
 
701
+ # 准备输出
702
  output_image = plot_detection(image, detection_result)
703
 
704
  if generate_tech_report:
705
+ # 准备技术报告的数据
706
  tech_report_data = {
707
  'model_outputs': model_outputs,
708
  'detection_results': detection_results,
 
710
  'performance_data': performance_data
711
  }
712
 
713
+ # 生成技术报告
714
  tech_report = TechnicalReportGenerator().generate_report(tech_report_data)
715
 
716
+ # 将技术报告保存到临时文件
717
  report_path = "temp_tech_report.md"
718
  with open(report_path, "w") as f:
719
  f.write(tech_report)
720
 
721
+ progress(1.0, desc="分析完成!")
722
+ # 处理完成后清理内存
723
+ cleanup_memory()
724
  return output_image, analysis_report, report_path, confidence_chart, feature_chart, heatmap
725
 
726
+ progress(1.0, desc="分析完成!")
727
+ # 处理完成后清理内存
728
+ cleanup_memory()
729
  return output_image, analysis_report, None, confidence_chart, feature_chart, heatmap
730
 
731
  except Exception as e:
732
+ error_msg = f"处理图像时出错: {str(e)}"
733
  print(error_msg)
734
+ # 出错后清理内存
735
+ cleanup_memory()
736
  raise gr.Error(error_msg)
737
 
738
  def display_history():
 
756
  def get_memory_usage():
757
  """Get current memory usage in MB"""
758
  process = psutil.Process()
759
+ memory_info = process.memory_info()
760
+ return memory_info.rss / 1024 / 1024
761
 
762
  def get_peak_memory_usage():
763
  """Get peak memory usage in MB"""
764
+ try:
765
+ process = psutil.Process()
766
+ memory_info = process.memory_info()
767
+ if hasattr(memory_info, 'peak_wset'):
768
+ return memory_info.peak_wset / 1024 / 1024
769
+ else:
770
+ # On Linux, we can use /proc/self/status to get peak memory
771
+ with open('/proc/self/status') as f:
772
+ for line in f:
773
+ if line.startswith('VmHWM:'):
774
+ return float(line.split()[1]) / 1024 # Convert KB to MB
775
+ except:
776
+ pass
777
+ return 0
778
 
779
  def get_gpu_utilization():
780
  """Get GPU utilization percentage"""
 
785
  pass
786
  return 0
787
 
788
+ def log_memory_usage(stage=""):
789
+ """Log current memory usage"""
790
+ mem_usage = get_memory_usage()
791
+ peak_mem = get_peak_memory_usage()
792
+ gpu_util = get_gpu_utilization()
793
+ print(f"Memory usage at {stage}: {mem_usage:.2f}MB (Peak: {peak_mem:.2f}MB, GPU: {gpu_util:.2f}%)")
794
+
795
  def toggle_dark_mode():
796
  """Toggle between light and dark themes"""
797
  global DARK_MODE
798
  DARK_MODE = not DARK_MODE
799
  return gr.Theme.darkmode() if DARK_MODE else THEME
800
 
801
+ def get_space_upgrade_url():
802
+ """Get the URL for upgrading the Space"""
803
+ if not is_running_in_space():
804
+ return "#"
805
+
806
+ space_id = os.environ.get("SPACE_ID", "")
807
+ if not space_id:
808
+ return "https://huggingface.co/pricing"
809
+
810
+ # Extract username and space name
811
+ parts = space_id.split("/")
812
+ if len(parts) != 2:
813
+ return "https://huggingface.co/pricing"
814
+
815
+ username, space_name = parts
816
+ return f"https://huggingface.co/spaces/{username}/{space_name}/settings"
817
+
818
+ def get_local_installation_instructions():
819
+ """Get instructions for running the app locally"""
820
+ required_memory = estimate_model_memory_requirements()
821
+ repo_url = get_repository_url()
822
+
823
+ return f"""
824
+ ## Running Locally
825
+
826
+ To run this application locally with the full model:
827
+
828
+ 1. Clone the repository:
829
+ ```bash
830
+ git clone {repo_url}
831
+ cd radar-analysis
832
+ ```
833
+
834
+ 2. Install dependencies:
835
+ ```bash
836
+ pip install -r requirements.txt
837
+ ```
838
+
839
+ 3. Set your Hugging Face token as an environment variable:
840
+ ```bash
841
+ export HF_TOCKEN=your_huggingface_token
842
+ ```
843
+
844
+ 4. Run the application:
845
+ ```bash
846
+ python app.py
847
+ ```
848
+
849
+ Make sure your system has at least {required_memory/1024:.1f}GB of RAM for optimal performance.
850
+ """
851
+
852
+ def get_model_card_url():
853
+ """Get the URL for the model card"""
854
+ return f"https://huggingface.co/{MODEL_NAME}"
855
+
856
+ def has_enough_memory_for_model():
857
+ """Check if we have enough memory for the model"""
858
+ if is_running_in_space():
859
+ # In Spaces, we need to be more cautious
860
+ hardware_memory = get_space_hardware_memory() * 1024 # Convert GB to MB
861
+ required_memory = estimate_model_memory_requirements()
862
+ print(f"Space hardware memory: {hardware_memory}MB, Required: {required_memory:.2f}MB")
863
+ return hardware_memory >= required_memory
864
+ else:
865
+ # For local development, check available memory
866
+ available_memory = check_available_memory()
867
+ required_memory = estimate_model_memory_requirements()
868
+ print(f"Available memory: {available_memory:.2f}MB, Required: {required_memory:.2f}MB")
869
+ return available_memory >= required_memory
870
+
871
+ def get_repository_url():
872
+ """Get the URL for the repository"""
873
+ if is_running_in_space():
874
+ space_id = os.environ.get("SPACE_ID", "")
875
+ if space_id:
876
+ # Space ID is in the format "username/spacename"
877
+ return f"https://huggingface.co/spaces/{space_id}"
878
+ else:
879
+ return "https://huggingface.co/spaces/xingqiang/radar-analysis"
880
+ else:
881
+ return "https://huggingface.co/spaces/xingqiang/radar-analysis"
882
+
883
+ def get_directory_name_from_repo_url(repo_url):
884
+ """Get the directory name from the repository URL"""
885
+ # Extract the last part of the URL
886
+ parts = repo_url.rstrip('/').split('/')
887
+ return parts[-1]
888
+
889
+ # Launch the interface
890
+ def launch():
891
+ """启动Gradio界面"""
892
+ if is_running_in_space():
893
+ # 在Spaces中,使用最小资源配置以避免内存问题
894
+ logger.info("使用最小资源配置启动Spaces")
895
+ iface.launch(
896
+ share=False,
897
+ server_name="0.0.0.0",
898
+ server_port=7860,
899
+ max_threads=4, # 从10减少到4
900
+ show_error=True,
901
+ quiet=False
902
+ )
903
+ else:
904
+ # 对于本地开发,使用默认设置
905
+ iface.launch()
906
+
907
  # Create Gradio interface
908
  with gr.Blocks(theme=THEME) as iface:
909
  theme_state = gr.State(THEME)
910
+
911
  with gr.Row():
912
+ gr.Markdown("# 雷达图像分析系统")
913
+ dark_mode_btn = gr.Button("🌓 切换暗黑模式", scale=0)
914
+
915
+ # 添加模型加载提示
916
+ gr.Markdown("""
917
+ ### ℹ️ 模型加载说明
918
+ - 模型仅在您点击"分析"按钮时才会下载和初始化
919
+ - 首次分析可能需要较长时间,因为需要下载模型
920
+ - 如果内存不足,系统会自动切换到演示模式
921
+ """, elem_id="model-loading-notice")
922
+
923
  if USE_DEMO_MODE:
924
+ hardware_type = get_space_hardware_type() if is_running_in_space() else "N/A"
925
+ hardware_tier = get_space_hardware_tier() if is_running_in_space() else "N/A"
926
+ hardware_memory = get_space_hardware_memory() if is_running_in_space() else 0
927
+ total_memory = get_total_system_memory()
928
+ required_memory = estimate_model_memory_requirements()
929
+ recommended_tier = get_recommended_space_tier()
930
+ upgrade_url = get_space_upgrade_url()
931
+ model_card_url = get_model_card_url()
932
+
933
+ memory_info = f"Space硬件: {hardware_type} (等级: {hardware_tier}, 内存: {hardware_memory}GB)"
934
+ model_req = f"[PaliGemma模型]({model_card_url})在使用8位量化加载时需要约{required_memory/1024:.1f}GB内存"
935
+
936
+ gr.Markdown(f"""
937
+ ### ⚠️ 运行在演示模式
938
+ 由于内存限制,应用程序当前在演示模式下运行:
939
+
940
+ 1. **内存错误**: Space遇到"内存限制超过(16Gi)"错误
941
+ - {memory_info}
942
+ - 系统总内存: {total_memory:.2f}MB
943
+ - {model_req}
944
+
945
+ 2. **解决方案**:
946
+ - 演示模式提供模拟结果用于演示目的
947
+ - 要使用完整模型,请在本地运行此应用程序,需要{required_memory/1024:.1f}GB+内存
948
+ - 或[升级到{recommended_tier} Space等级]({upgrade_url})或更高
949
+
950
+ 演示模式仍提供所有UI功能和可视化特性。
951
  """, elem_id="demo-mode-warning")
952
+
953
+ gr.Markdown("上传雷达图像以分析缺陷并生成技术报告")
954
+
955
  with gr.Tabs() as tabs:
956
+ with gr.TabItem("分析", id="analysis"):
957
  with gr.Row():
958
  with gr.Column(scale=1):
959
+ with gr.Accordion("输入", open=True):
960
  input_image = gr.Image(
961
+ type="pil",
962
+ label="上传雷达图像",
963
  elem_id="input-image",
964
  sources=["upload", "webcam", "clipboard"],
965
  tool="editor"
966
  )
967
  tech_report_checkbox = gr.Checkbox(
968
+ label="生成技术报告",
969
  value=False,
970
+ info="创建详细的技术分析报告"
971
  )
972
  analyze_button = gr.Button(
973
+ "分析",
974
  variant="primary",
975
  elem_id="analyze-btn"
976
  )
977
+
978
  with gr.Column(scale=2):
979
+ with gr.Accordion("检测结果", open=True):
980
  output_image = gr.Image(
981
+ type="pil",
982
+ label="检测结果",
983
  elem_id="output-image"
984
  )
985
+
986
+ with gr.Accordion("分析报告", open=True):
987
  output_report = gr.HTML(
988
+ label="分析报告",
989
  elem_id="analysis-report"
990
  )
991
  tech_report_output = gr.File(
992
+ label="技术报告",
993
  elem_id="tech-report"
994
  )
995
+
996
  with gr.Row():
997
  with gr.Column():
998
  confidence_plot = gr.Plot(
999
+ label="置信度分数",
1000
  elem_id="confidence-plot"
1001
  )
1002
+
1003
  with gr.Column():
1004
  feature_plot = gr.Plot(
1005
+ label="特征分析",
1006
  elem_id="feature-plot"
1007
  )
1008
+
1009
  with gr.Row():
1010
  heatmap_plot = gr.Plot(
1011
+ label="信号强度热图",
1012
  elem_id="heatmap-plot"
1013
  )
1014
+
1015
+ with gr.TabItem("历史", id="history"):
1016
  with gr.Row():
1017
+ history_button = gr.Button("刷新历史")
1018
  history_output = gr.HTML(elem_id="history-output")
1019
+
1020
+ with gr.TabItem("帮助", id="help"):
1021
  gr.Markdown("""
1022
+ ## 如何使用此工具
1023
+
1024
+ 1. **上传图像**: 点击上传按钮选择要分析的雷达图像
1025
+ 2. **生成技术报告** (可选): 如果需要详细的技术报告,请勾选此框
1026
+ 3. **分析**: 点击分析按钮处理图像
1027
+ 4. **查看结果**:
1028
+ - 检测可视化显示已识别的缺陷
1029
+ - 分析报告提供发现的摘要
1030
+ - 技术报告(如果请求)提供详细指标
1031
+ - 图表提供置信度分数和特征分析的可视化表示
1032
+
1033
+ ## 关于模型
1034
+
1035
+ 该系统使用[PaliGemma]({get_model_card_url()}),这是一个视觉-语言模型,结合了SigLIP-So400m(图像编码器)Gemma-2B(文本解码器)进行联合目标检测和多模态分析。
1036
+
1037
+ 该模型针对雷达图像分析进行了微调,可以检测结构检查图像中的各种类型的缺陷和异常。
 
 
 
 
 
 
 
 
 
1038
  """)
1039
+
1040
+ if USE_DEMO_MODE and is_running_in_space():
1041
+ gr.Markdown(get_local_installation_instructions())
1042
+
1043
+ gr.Markdown("""
1044
+ ## 键盘快捷键
1045
+
1046
+ - **Ctrl+A**: 触发分析
1047
+ - **Ctrl+D**: 切换暗黑模式
1048
+
1049
+ ## 故障排除
1050
+
1051
+ - 如果分析失败,请尝试上传不同的图像格式
1052
+ - 确保图像是有效的雷达扫描
1053
+ - 对于技术问题,请查看控制台日志
1054
+ """)
1055
+
1056
  # Set up event handlers
1057
  dark_mode_btn.click(
1058
  fn=toggle_dark_mode,
 
1060
  outputs=[iface],
1061
  api_name="toggle_theme"
1062
  )
1063
+
1064
  analyze_button.click(
1065
  fn=process_image_streaming,
1066
  inputs=[input_image, tech_report_checkbox],
1067
  outputs=[output_image, output_report, tech_report_output, confidence_plot, feature_plot, heatmap_plot],
1068
  api_name="analyze"
1069
  )
1070
+
1071
  history_button.click(
1072
  fn=display_history,
1073
  inputs=[],
1074
  outputs=[history_output],
1075
  api_name="history"
1076
  )
1077
+
1078
  # Add keyboard shortcuts
1079
  iface.load(lambda: None, None, None, _js="""
1080
  () => {
 
1083
  document.getElementById('analyze-btn').click();
1084
  }
1085
  if (e.key === 'd' && e.ctrlKey) {
1086
+ document.querySelector('button:contains("切换暗黑模式")').click();
1087
  }
1088
  });
1089
  }
1090
  """)
1091
 
1092
  # Launch the interface
1093
+ launch()
config.py CHANGED
@@ -2,7 +2,7 @@ import os
2
 
3
  BASE_DIR = os.path.dirname(os.path.abspath(__file__))
4
 
5
- MODEL_NAME = "Extremely4606/paligemma_9_19"
6
  DATABASE_URL = f"sqlite:///{os.path.join(BASE_DIR, 'radar_reports.db')}"
7
 
8
  AMPLITUDE_THRESHOLD = 128
 
2
 
3
  BASE_DIR = os.path.dirname(os.path.abspath(__file__))
4
 
5
+ MODEL_NAME = "Extremely4606/paligemma24_12_30"
6
  DATABASE_URL = f"sqlite:///{os.path.join(BASE_DIR, 'radar_reports.db')}"
7
 
8
  AMPLITUDE_THRESHOLD = 128
create_space.py CHANGED
@@ -4,13 +4,23 @@ import sys
4
 
5
  def create_and_push_space():
6
  try:
 
 
 
 
 
 
 
 
7
  print("Creating Space...")
8
  # Create the space
 
9
  repo_url = create_repo(
10
  repo_id="xingqiang/radar-analysis",
11
  repo_type="space",
12
  space_sdk="gradio",
13
- private=False
 
14
  )
15
  print(f"Space created successfully at: {repo_url}")
16
 
 
4
 
5
  def create_and_push_space():
6
  try:
7
+ # Get Hugging Face token from environment
8
+ hf_token = os.environ.get("HF_TOKEN")
9
+ if not hf_token:
10
+ print("Error: HF_TOKEN environment variable not set")
11
+ print("Please set your Hugging Face token as an environment variable:")
12
+ print("export HF_TOKEN=your_token_here")
13
+ sys.exit(1)
14
+
15
  print("Creating Space...")
16
  # Create the space
17
+ api = HfApi(token=hf_token)
18
  repo_url = create_repo(
19
  repo_id="xingqiang/radar-analysis",
20
  repo_type="space",
21
  space_sdk="gradio",
22
+ private=False,
23
+ token=hf_token
24
  )
25
  print(f"Space created successfully at: {repo_url}")
26
 
feature_extraction.py CHANGED
@@ -40,13 +40,30 @@ def classify_reflections(count):
40
 
41
 
42
  def extract_features(image, detection_result):
 
 
 
 
 
 
 
 
 
 
43
  np_image = np.array(image)
44
  amplitude = calculate_amplitude(np_image)
45
  amplitude_class = classify_amplitude(amplitude)
46
 
47
- box = detection_result['boxes'][0].tolist()
48
- distribution_range = calculate_distribution_range(box)
49
- distribution_class = classify_distribution_range(distribution_range)
 
 
 
 
 
 
 
50
 
51
  attenuation_rate = calculate_attenuation_rate(np_image)
52
  attenuation_class = classify_attenuation_rate(attenuation_rate)
@@ -59,8 +76,8 @@ def extract_features(image, detection_result):
59
  "分布范围": distribution_class,
60
  "衰减速度": attenuation_class,
61
  "反射次数": reflection_class,
62
- "振幅值": amplitude,
63
- "分布范围值": distribution_range,
64
- "衰减速度值": attenuation_rate,
65
- "反射次数值": reflection_count
66
  }
 
40
 
41
 
42
  def extract_features(image, detection_result):
43
+ """
44
+ Extract features from the image and detection result.
45
+
46
+ Args:
47
+ image: PIL Image
48
+ detection_result: Dictionary containing detection results
49
+
50
+ Returns:
51
+ Dictionary of features
52
+ """
53
  np_image = np.array(image)
54
  amplitude = calculate_amplitude(np_image)
55
  amplitude_class = classify_amplitude(amplitude)
56
 
57
+ # Handle box calculation
58
+ if detection_result and 'boxes' in detection_result and len(detection_result['boxes']) > 0:
59
+ box = detection_result['boxes'][0]
60
+ if not isinstance(box, list):
61
+ box = box.tolist()
62
+ distribution_range = calculate_distribution_range(box)
63
+ distribution_class = classify_distribution_range(distribution_range)
64
+ else:
65
+ distribution_range = 0
66
+ distribution_class = "小"
67
 
68
  attenuation_rate = calculate_attenuation_rate(np_image)
69
  attenuation_class = classify_attenuation_rate(attenuation_rate)
 
76
  "分布范围": distribution_class,
77
  "衰减速度": attenuation_class,
78
  "反射次数": reflection_class,
79
+ "振幅值": float(amplitude),
80
+ "分布范围值": float(distribution_range),
81
+ "衰减速度值": float(attenuation_rate),
82
+ "反射次数值": int(reflection_count)
83
  }
model.py CHANGED
@@ -6,101 +6,205 @@ import logging
6
  from transformers import AutoProcessor, AutoModelForVision2Seq
7
  from PIL import Image
8
  import numpy as np
 
9
 
 
 
 
10
  logger = logging.getLogger(__name__)
11
 
12
  class RadarDetectionModel:
13
- def __init__(self, model_name="Extremely4606/paligemma24_12_30", use_auth_token=None):
14
  """
15
- Initialize the radar detection model.
16
 
17
  Args:
18
- model_name (str): The name or path of the model to load
19
- use_auth_token (str, optional): Hugging Face token for accessing gated models
20
  """
21
  self.device = "cuda" if torch.cuda.is_available() else "cpu"
 
22
 
23
- # Load model and processor
24
- if use_auth_token:
25
- self.processor = AutoProcessor.from_pretrained(model_name, use_auth_token=use_auth_token)
26
- self.model = AutoModelForVision2Seq.from_pretrained(model_name, use_auth_token=use_auth_token)
 
 
27
  else:
28
- self.processor = AutoProcessor.from_pretrained(model_name)
29
- self.model = AutoModelForVision2Seq.from_pretrained(model_name)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
30
 
31
- self.model.to(self.device)
32
- self.model.eval()
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
33
 
34
  def detect(self, image):
35
  """
36
- Detect objects in the radar image.
37
 
38
  Args:
39
- image (PIL.Image): The radar image to analyze
40
 
41
  Returns:
42
- dict: Detection results including boxes, scores, and labels
43
  """
44
- # Preprocess image
45
- inputs = self.processor(images=image, return_tensors="pt").to(self.device)
46
-
47
- # Run inference
48
- with torch.no_grad():
49
- outputs = self.model.generate(
50
- **inputs,
51
- max_length=50,
52
- num_beams=4,
53
- early_stopping=True
54
- )
55
-
56
- # Process outputs
57
- generated_text = self.processor.batch_decode(outputs, skip_special_tokens=True)[0]
58
-
59
- # Parse detection results from generated text
60
- # This is a simplified example - actual parsing would depend on model output format
61
- boxes, scores, labels = self._parse_detection_results(generated_text, image.size)
62
-
63
- return {
64
- 'boxes': boxes,
65
- 'scores': scores,
66
- 'labels': labels,
67
- 'image': image
68
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
69
 
70
  def _parse_detection_results(self, text, image_size):
71
  """
72
- Parse detection results from generated text.
73
 
74
  Args:
75
- text (str): Generated text from the model
76
- image_size (tuple): Size of the input image (width, height)
77
 
78
  Returns:
79
  tuple: (boxes, scores, labels)
80
  """
81
- # This is a simplified example - actual parsing would depend on model output format
82
- # For demonstration, we'll extract some mock detections
83
 
84
- # Check for common defect keywords in the text
85
  defects = []
86
 
87
- if "crack" in text.lower():
88
- defects.append(("Crack", 0.92, [0.2, 0.3, 0.4, 0.5]))
89
 
90
- if "corrosion" in text.lower():
91
- defects.append(("Corrosion", 0.85, [0.6, 0.2, 0.8, 0.4]))
92
 
93
- if "damage" in text.lower():
94
- defects.append(("Damage", 0.78, [0.1, 0.7, 0.3, 0.9]))
95
 
96
- if "defect" in text.lower():
97
- defects.append(("Defect", 0.88, [0.5, 0.5, 0.7, 0.7]))
98
 
99
- # If no defects found, add a generic one
100
  if not defects:
101
- defects.append(("Anomaly", 0.75, [0.4, 0.4, 0.6, 0.6]))
102
 
103
- # Convert normalized coordinates to pixel coordinates
104
  width, height = image_size
105
  boxes = []
106
  scores = []
 
6
  from transformers import AutoProcessor, AutoModelForVision2Seq
7
  from PIL import Image
8
  import numpy as np
9
+ from config import MODEL_NAME
10
 
11
+ # 配置日志记录
12
+ logging.basicConfig(level=logging.INFO,
13
+ format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
14
  logger = logging.getLogger(__name__)
15
 
16
  class RadarDetectionModel:
17
+ def __init__(self, model_name=None, use_auth_token=None):
18
  """
19
+ 初始化雷达检测模型。
20
 
21
  Args:
22
+ model_name (str): 要加载的模型名称或路径
23
+ use_auth_token (str, optional): 用于访问受限模型的Hugging Face令牌
24
  """
25
  self.device = "cuda" if torch.cuda.is_available() else "cpu"
26
+ logger.info(f"使用设备: {self.device}")
27
 
28
+ self.model_name = model_name if model_name else MODEL_NAME
29
+ logger.info(f"模型名称: {self.model_name}")
30
+
31
+ self.use_auth_token = use_auth_token or os.environ.get("HF_TOKEN")
32
+ if self.use_auth_token:
33
+ logger.info("已提供Hugging Face令牌")
34
  else:
35
+ logger.warning("未提供Hugging Face令牌,可能无法访问受限模型")
36
+
37
+ self.processor = None
38
+ self.model = None
39
+
40
+ # 加载模型和处理器
41
+ logger.info("开始加载模型和处理器...")
42
+ self._load_model()
43
+
44
+ def _load_model(self):
45
+ """加载模型和处理器,并监控内存使用情况"""
46
+ try:
47
+ logger.info(f"正在从{self.model_name}加载处理器")
48
+ start_time = torch.cuda.Event(enable_timing=True) if torch.cuda.is_available() else None
49
+ end_time = torch.cuda.Event(enable_timing=True) if torch.cuda.is_available() else None
50
+
51
+ if start_time:
52
+ start_time.record()
53
+
54
+ if self.use_auth_token:
55
+ # 如果提供了令牌,登录到Hugging Face Hub
56
+ logger.info("使用令牌登录到Hugging Face Hub")
57
+ login(token=self.use_auth_token)
58
+ self.processor = AutoProcessor.from_pretrained(self.model_name, use_auth_token=self.use_auth_token)
59
+ else:
60
+ self.processor = AutoProcessor.from_pretrained(self.model_name)
61
+
62
+ if end_time:
63
+ end_time.record()
64
+ torch.cuda.synchronize()
65
+ logger.info(f"处理器加载时间: {start_time.elapsed_time(end_time):.2f}毫秒")
66
+
67
+ logger.info(f"正在从{self.model_name}加载模型,使用8位量化以减少内存使用")
68
 
69
+ if start_time:
70
+ start_time.record()
71
+
72
+ # 使用8位量化以减少内存使用
73
+ if self.use_auth_token:
74
+ self.model = AutoModelForVision2Seq.from_pretrained(
75
+ self.model_name,
76
+ use_auth_token=self.use_auth_token,
77
+ load_in_8bit=True, # 使用8位量化
78
+ device_map="auto" # 自动管理设备放置
79
+ )
80
+ else:
81
+ self.model = AutoModelForVision2Seq.from_pretrained(
82
+ self.model_name,
83
+ load_in_8bit=True, # 使用8位量化
84
+ device_map="auto" # 自动管理设备放置
85
+ )
86
+
87
+ if end_time:
88
+ end_time.record()
89
+ torch.cuda.synchronize()
90
+ logger.info(f"模型加载时间: {start_time.elapsed_time(end_time):.2f}毫秒")
91
+
92
+ logger.info(f"模型加载成功")
93
+ # 使用device_map="auto"时无需手动移动到设备
94
+ self.model.eval()
95
+
96
+ # 记录模型信息
97
+ param_count = sum(p.numel() for p in self.model.parameters())
98
+ logger.info(f"模型参数数量: {param_count:,}")
99
+
100
+ if torch.cuda.is_available():
101
+ memory_allocated = torch.cuda.memory_allocated() / (1024 * 1024)
102
+ memory_reserved = torch.cuda.memory_reserved() / (1024 * 1024)
103
+ logger.info(f"GPU内存分配: {memory_allocated:.2f}MB")
104
+ logger.info(f"GPU内存保留: {memory_reserved:.2f}MB")
105
+
106
+ except Exception as e:
107
+ logger.error(f"加载模型时出错: {str(e)}")
108
+ raise
109
 
110
  def detect(self, image):
111
  """
112
+ 检测雷达图像中的对象。
113
 
114
  Args:
115
+ image (PIL.Image): 要分析的雷达图像
116
 
117
  Returns:
118
+ dict: 检测结果,包括边界框、分数和标签
119
  """
120
+ try:
121
+ if self.model is None or self.processor is None:
122
+ raise ValueError("模型或处理器未正确初始化")
123
+
124
+ # 预处理图像
125
+ logger.info("预处理图像")
126
+ inputs = self.processor(images=image, return_tensors="pt").to(self.device)
127
+
128
+ # 运行推理
129
+ logger.info("运行模型推理")
130
+ start_time = torch.cuda.Event(enable_timing=True) if torch.cuda.is_available() else None
131
+ end_time = torch.cuda.Event(enable_timing=True) if torch.cuda.is_available() else None
132
+
133
+ if start_time:
134
+ start_time.record()
135
+
136
+ with torch.no_grad():
137
+ outputs = self.model.generate(
138
+ **inputs,
139
+ max_length=50,
140
+ num_beams=4,
141
+ early_stopping=True
142
+ )
143
+
144
+ if end_time:
145
+ end_time.record()
146
+ torch.cuda.synchronize()
147
+ inference_time = start_time.elapsed_time(end_time)
148
+ logger.info(f"推理时间: {inference_time:.2f}毫秒")
149
+
150
+ # 处理输出
151
+ generated_text = self.processor.batch_decode(outputs, skip_special_tokens=True)[0]
152
+ logger.info(f"生成的文本: {generated_text}")
153
+
154
+ # 从生成的文本中解析检测结果
155
+ boxes, scores, labels = self._parse_detection_results(generated_text, image.size)
156
+ logger.info(f"检测到{len(boxes)}个对象")
157
+
158
+ return {
159
+ 'boxes': boxes,
160
+ 'scores': scores,
161
+ 'labels': labels,
162
+ 'image': image
163
+ }
164
+ except Exception as e:
165
+ logger.error(f"检测过程中出错: {str(e)}")
166
+ # 返回备用检测结果
167
+ return {
168
+ 'boxes': [[100, 100, 200, 200]],
169
+ 'scores': [0.75],
170
+ 'labels': ['错误: ' + str(e)[:50]],
171
+ 'image': image
172
+ }
173
 
174
  def _parse_detection_results(self, text, image_size):
175
  """
176
+ 从生成的文本中解析检测结果。
177
 
178
  Args:
179
+ text (str): 模型生成的文本
180
+ image_size (tuple): 输入图像的大小(宽度, 高度)
181
 
182
  Returns:
183
  tuple: (boxes, scores, labels)
184
  """
185
+ # 这是一个简化的示例 - 实际解析将取决于模型输出格式
186
+ # 为了演示,我们将提取一些模拟检测结果
187
 
188
+ # 检查文本中常见的缺陷关键词
189
  defects = []
190
 
191
+ if "crack" in text.lower() or "裂缝" in text.lower():
192
+ defects.append(("裂缝", 0.92, [0.2, 0.3, 0.4, 0.5]))
193
 
194
+ if "corrosion" in text.lower() or "腐蚀" in text.lower():
195
+ defects.append(("腐蚀", 0.85, [0.6, 0.2, 0.8, 0.4]))
196
 
197
+ if "damage" in text.lower() or "损坏" in text.lower():
198
+ defects.append(("损坏", 0.78, [0.1, 0.7, 0.3, 0.9]))
199
 
200
+ if "defect" in text.lower() or "缺陷" in text.lower():
201
+ defects.append(("缺陷", 0.88, [0.5, 0.5, 0.7, 0.7]))
202
 
203
+ # 如果没有找到缺陷,添加一个通用的
204
  if not defects:
205
+ defects.append(("异常", 0.75, [0.4, 0.4, 0.6, 0.6]))
206
 
207
+ # 将归一化坐标转换为像素坐标
208
  width, height = image_size
209
  boxes = []
210
  scores = []
requirements.txt CHANGED
@@ -1,22 +1,24 @@
1
- gradio>=5.18.0
2
- torch>=2.1.2
3
- transformers>=4.37.2
4
- Pillow>=10.2.0
5
- numpy>=1.26.3
6
  matplotlib>=3.8.2
7
- pandas>=2.1.4
8
  sqlalchemy>=2.0.25
9
- plotly>=5.18.0
10
  scikit-learn>=1.3.2
11
  jinja2>=3.1.3
12
- huggingface-hub>=0.20.2
13
  python-dotenv>=1.0.0
14
  markdown>=3.5.1
15
- psutil>=5.9.6
16
  tqdm>=4.66.1
17
- accelerate>=0.25.0
18
  safetensors>=0.4.1
19
  peft>=0.7.1
20
  optimum>=1.14.0
21
  colorama>=0.4.6
22
- rich>=13.7.0
 
 
 
1
+ gradio==4.19.2
2
+ torch==2.1.2
3
+ transformers==4.37.2
4
+ pillow==10.1.0
5
+ numpy==1.26.2
6
  matplotlib>=3.8.2
7
+ pandas==2.1.3
8
  sqlalchemy>=2.0.25
9
+ plotly==5.18.0
10
  scikit-learn>=1.3.2
11
  jinja2>=3.1.3
12
+ huggingface-hub==0.20.2
13
  python-dotenv>=1.0.0
14
  markdown>=3.5.1
15
+ psutil==5.9.6
16
  tqdm>=4.66.1
17
+ accelerate==0.25.0
18
  safetensors>=0.4.1
19
  peft>=0.7.1
20
  optimum>=1.14.0
21
  colorama>=0.4.6
22
+ rich>=13.7.0
23
+ bitsandbytes==0.41.1
24
+ scipy>=1.11.3
run.py ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import os
2
+ import sys
3
+ import logging
4
+ from dotenv import load_dotenv
5
+
6
+ # Load environment variables from .env file if it exists
7
+ load_dotenv()
8
+
9
+ # Configure logging
10
+ logging.basicConfig(level=logging.INFO,
11
+ format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
12
+ logger = logging.getLogger(__name__)
13
+
14
+ def main():
15
+ """Run the Radar Analysis System application"""
16
+ try:
17
+ # Check for HF_TOKEN environment variable
18
+ hf_token = os.environ.get("HF_TOKEN")
19
+ if not hf_token:
20
+ logger.warning("HF_TOKEN environment variable not set. The application will run in demo mode.")
21
+ else:
22
+ logger.info("HF_TOKEN environment variable found.")
23
+
24
+ # Import app module
25
+ import app
26
+
27
+ # Launch the application
28
+ logger.info("Starting Radar Analysis System...")
29
+ app.launch()
30
+
31
+ return True
32
+ except Exception as e:
33
+ logger.error(f"Error running application: {str(e)}")
34
+ return False
35
+
36
+ if __name__ == "__main__":
37
+ success = main()
38
+ sys.exit(0 if success else 1)
test_app.py ADDED
@@ -0,0 +1,129 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import os
2
+ import sys
3
+ import logging
4
+
5
+ # Configure logging
6
+ logging.basicConfig(level=logging.INFO,
7
+ format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
8
+ logger = logging.getLogger(__name__)
9
+
10
+ def test_imports():
11
+ """测试所有必需的模块都可以导入"""
12
+ try:
13
+ import torch
14
+ logger.info(f"PyTorch版本: {torch.__version__}")
15
+
16
+ import transformers
17
+ logger.info(f"Transformers版本: {transformers.__version__}")
18
+
19
+ import numpy as np
20
+ logger.info(f"NumPy版本: {np.__version__}")
21
+
22
+ import PIL
23
+ logger.info(f"PIL版本: {PIL.__version__}")
24
+
25
+ import scipy
26
+ logger.info(f"SciPy版本: {scipy.__version__}")
27
+
28
+ logger.info("所有导入成功")
29
+ return True
30
+ except ImportError as e:
31
+ logger.error(f"导入错误: {str(e)}")
32
+ return False
33
+
34
+ def test_model_loading():
35
+ """测试模型可以加载"""
36
+ try:
37
+ from model import RadarDetectionModel
38
+
39
+ # 检查是否设置了HF_TOKEN环境变量
40
+ hf_token = os.environ.get("HF_TOKEN")
41
+ if not hf_token:
42
+ logger.warning("未设置HF_TOKEN环境变量,使用公共模型进行测试")
43
+
44
+ # 尝试初始化模型,使用较小的公共模型
45
+ logger.info("尝试初始化模型(使用较小的公共模型)")
46
+ model = RadarDetectionModel(model_name="google/siglip-base-patch16-224")
47
+ logger.info("模型初始化成功")
48
+ return True
49
+ except Exception as e:
50
+ logger.error(f"模型加载错误: {str(e)}")
51
+ return False
52
+
53
+ def test_feature_extraction():
54
+ """测试特征提取功能"""
55
+ try:
56
+ import numpy as np
57
+ from PIL import Image
58
+ from feature_extraction import extract_features
59
+
60
+ # 创建一个虚拟图像和检测结果
61
+ logger.info("创建虚拟测试数据")
62
+ dummy_image = Image.fromarray(np.random.randint(0, 255, (224, 224, 3), dtype=np.uint8))
63
+ dummy_detection = {
64
+ 'boxes': [[50, 50, 100, 100]],
65
+ 'scores': [0.9],
66
+ 'labels': ['测试']
67
+ }
68
+
69
+ # 提取特征
70
+ logger.info("提取特征")
71
+ features = extract_features(dummy_image, dummy_detection)
72
+ logger.info(f"提取的特征: {features}")
73
+ return True
74
+ except Exception as e:
75
+ logger.error(f"特征提取错误: {str(e)}")
76
+ return False
77
+
78
+ def test_app_initialization():
79
+ """测试应用程序初始化但不加载模型"""
80
+ try:
81
+ logger.info("测试应用程序初始化")
82
+ import app
83
+
84
+ # 检查应用程序是否已初始化但没有加载模型
85
+ logger.info("检查应用程序全局变量")
86
+ assert app.model is None, "模型不应该在导入时加载"
87
+ assert app.MODEL_INIT_ATTEMPTED is False, "模型初始化尝试标志应为False"
88
+
89
+ logger.info("应用程序初始化测试通过")
90
+ return True
91
+ except Exception as e:
92
+ logger.error(f"应用程序初始化错误: {str(e)}")
93
+ return False
94
+
95
+ def run_tests():
96
+ """运行所有测试"""
97
+ tests = [
98
+ ("导入测试", test_imports),
99
+ ("应用程序初始化测试", test_app_initialization),
100
+ ("模型加载测试", test_model_loading),
101
+ ("特征提取测试", test_feature_extraction)
102
+ ]
103
+
104
+ results = []
105
+ for name, test_func in tests:
106
+ logger.info(f"运行{name}...")
107
+ try:
108
+ result = test_func()
109
+ results.append((name, result))
110
+ logger.info(f"{name}: {'通过' if result else '失败'}")
111
+ except Exception as e:
112
+ logger.error(f"{name}失败,错误: {str(e)}")
113
+ results.append((name, False))
114
+
115
+ # 打印摘要
116
+ logger.info("\n--- 测试摘要 ---")
117
+ passed = sum(1 for _, result in results if result)
118
+ total = len(results)
119
+ logger.info(f"通过: {passed}/{total} 测试")
120
+
121
+ for name, result in results:
122
+ status = "通过" if result else "失败"
123
+ logger.info(f"{name}: {status}")
124
+
125
+ return passed == total
126
+
127
+ if __name__ == "__main__":
128
+ success = run_tests()
129
+ sys.exit(0 if success else 1)
utils.py CHANGED
@@ -9,10 +9,12 @@ def plot_detection(image, detection_result):
9
  ax = plt.gca()
10
 
11
  for score, label, box in zip(detection_result["scores"], detection_result["labels"], detection_result["boxes"]):
12
- x, y, w, h = box
13
- rect = plt.Rectangle((x, y), w-x, h-y, fill=False, color='red')
 
 
14
  ax.add_patch(rect)
15
- ax.text(x, y, f'{label}: {score:.2f}',
16
  bbox=dict(facecolor='white', alpha=0.8))
17
 
18
  plt.axis('off')
 
9
  ax = plt.gca()
10
 
11
  for score, label, box in zip(detection_result["scores"], detection_result["labels"], detection_result["boxes"]):
12
+ x1, y1, x2, y2 = box
13
+ width = x2 - x1
14
+ height = y2 - y1
15
+ rect = plt.Rectangle((x1, y1), width, height, fill=False, color='red', linewidth=2)
16
  ax.add_patch(rect)
17
+ ax.text(x1, y1, f'{label}: {score:.2f}',
18
  bbox=dict(facecolor='white', alpha=0.8))
19
 
20
  plt.axis('off')