Spaces:
Sleeping
Sleeping
Dileep7729
commited on
Update app.py
Browse files
app.py
CHANGED
@@ -1,82 +1,39 @@
|
|
1 |
import gradio as gr
|
2 |
from transformers import CLIPModel, CLIPProcessor
|
3 |
from PIL import Image
|
4 |
-
import requests
|
5 |
|
6 |
-
#
|
7 |
model_name = "quadranttechnologies/retail-content-safety-clip-finetuned"
|
8 |
-
|
9 |
print("Initializing the application...")
|
10 |
|
11 |
try:
|
12 |
-
print("Loading the model from Hugging Face Model Hub...")
|
13 |
model = CLIPModel.from_pretrained(model_name, trust_remote_code=True)
|
14 |
processor = CLIPProcessor.from_pretrained(model_name)
|
15 |
print("Model and processor loaded successfully.")
|
16 |
except Exception as e:
|
17 |
print(f"Error loading the model or processor: {e}")
|
18 |
-
raise
|
19 |
-
|
20 |
-
# Step 2: Minimal Test Case to Verify Model and Processor
|
21 |
-
try:
|
22 |
-
print("Running a minimal test case with the model...")
|
23 |
-
|
24 |
-
# Test Image URL
|
25 |
-
url = "https://huggingface.co/datasets/Narsil/image_dummy/raw/main/parrots.png"
|
26 |
-
image = Image.open(requests.get(url, stream=True).raw)
|
27 |
-
|
28 |
-
# Define test categories
|
29 |
-
test_categories = ["safe", "unsafe"]
|
30 |
-
|
31 |
-
# Process the image
|
32 |
-
test_inputs = processor(text=test_categories, images=image, return_tensors="pt", padding=True)
|
33 |
-
print(f"Test inputs processed: {test_inputs}")
|
34 |
|
35 |
-
|
36 |
-
test_outputs = model(**test_inputs)
|
37 |
-
print(f"Test outputs: {test_outputs}")
|
38 |
-
|
39 |
-
# Check probabilities
|
40 |
-
test_logits = test_outputs.logits_per_image
|
41 |
-
test_probs = test_logits.softmax(dim=1)
|
42 |
-
print(f"Test probabilities: {test_probs}")
|
43 |
-
|
44 |
-
except Exception as e:
|
45 |
-
print(f"Error during the minimal test case: {e}")
|
46 |
-
raise RuntimeError(f"Test case failed: {e}")
|
47 |
-
|
48 |
-
# Step 3: Define the Inference Function
|
49 |
def classify_image(image):
|
50 |
-
"""
|
51 |
-
Classify an image as 'safe' or 'unsafe' and return probabilities.
|
52 |
-
|
53 |
-
Args:
|
54 |
-
image (PIL.Image.Image): Uploaded image.
|
55 |
-
|
56 |
-
Returns:
|
57 |
-
str: Predicted category.
|
58 |
-
dict: Probabilities for "safe" and "unsafe".
|
59 |
-
"""
|
60 |
try:
|
61 |
print("Starting image classification...")
|
62 |
|
63 |
-
#
|
64 |
if image is None:
|
65 |
raise ValueError("No image provided. Please upload a valid image.")
|
66 |
if not hasattr(image, "convert"):
|
67 |
raise ValueError("Uploaded file is not a valid image format.")
|
68 |
|
69 |
-
# Define
|
70 |
categories = ["safe", "unsafe"]
|
71 |
print(f"Categories: {categories}")
|
72 |
|
73 |
# Process the image
|
74 |
-
print("Processing the image with the processor...")
|
75 |
inputs = processor(text=categories, images=image, return_tensors="pt", padding=True)
|
76 |
print(f"Processed inputs: {inputs}")
|
77 |
|
78 |
# Perform inference
|
79 |
-
print("Running model inference...")
|
80 |
outputs = model(**inputs)
|
81 |
print(f"Model outputs: {outputs}")
|
82 |
|
@@ -85,34 +42,31 @@ def classify_image(image):
|
|
85 |
probs = logits_per_image.softmax(dim=1)
|
86 |
print(f"Probabilities: {probs}")
|
87 |
|
88 |
-
# Extract probabilities
|
89 |
safe_prob = probs[0][0].item() * 100
|
90 |
unsafe_prob = probs[0][1].item() * 100
|
|
|
91 |
|
92 |
# Determine the predicted category
|
93 |
predicted_category = "safe" if safe_prob > unsafe_prob else "unsafe"
|
94 |
-
print(f"Predicted category: {predicted_category}")
|
95 |
-
|
96 |
-
# Return the predicted category and probabilities
|
97 |
return predicted_category, {"safe": f"{safe_prob:.2f}%", "unsafe": f"{unsafe_prob:.2f}%"}
|
98 |
|
99 |
except Exception as e:
|
100 |
print(f"Error during classification: {e}")
|
101 |
return f"Error: {str(e)}", {}
|
102 |
|
103 |
-
#
|
104 |
iface = gr.Interface(
|
105 |
fn=classify_image,
|
106 |
inputs=gr.Image(type="pil"),
|
107 |
outputs=[
|
108 |
-
gr.Textbox(label="Predicted Category"),
|
109 |
-
gr.Label(label="Probabilities"),
|
110 |
],
|
111 |
title="Content Safety Classification",
|
112 |
description="Upload an image to classify it as 'safe' or 'unsafe' with corresponding probabilities.",
|
113 |
)
|
114 |
|
115 |
-
# Step 5: Launch Gradio Interface
|
116 |
if __name__ == "__main__":
|
117 |
print("Launching Gradio interface...")
|
118 |
iface.launch()
|
@@ -137,5 +91,6 @@ if __name__ == "__main__":
|
|
137 |
|
138 |
|
139 |
|
|
|
140 |
|
141 |
|
|
|
1 |
import gradio as gr
|
2 |
from transformers import CLIPModel, CLIPProcessor
|
3 |
from PIL import Image
|
|
|
4 |
|
5 |
+
# Load the model and processor
|
6 |
model_name = "quadranttechnologies/retail-content-safety-clip-finetuned"
|
|
|
7 |
print("Initializing the application...")
|
8 |
|
9 |
try:
|
|
|
10 |
model = CLIPModel.from_pretrained(model_name, trust_remote_code=True)
|
11 |
processor = CLIPProcessor.from_pretrained(model_name)
|
12 |
print("Model and processor loaded successfully.")
|
13 |
except Exception as e:
|
14 |
print(f"Error loading the model or processor: {e}")
|
15 |
+
raise
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
|
17 |
+
# Define the inference function
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
def classify_image(image):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
try:
|
20 |
print("Starting image classification...")
|
21 |
|
22 |
+
# Validate image input
|
23 |
if image is None:
|
24 |
raise ValueError("No image provided. Please upload a valid image.")
|
25 |
if not hasattr(image, "convert"):
|
26 |
raise ValueError("Uploaded file is not a valid image format.")
|
27 |
|
28 |
+
# Define categories
|
29 |
categories = ["safe", "unsafe"]
|
30 |
print(f"Categories: {categories}")
|
31 |
|
32 |
# Process the image
|
|
|
33 |
inputs = processor(text=categories, images=image, return_tensors="pt", padding=True)
|
34 |
print(f"Processed inputs: {inputs}")
|
35 |
|
36 |
# Perform inference
|
|
|
37 |
outputs = model(**inputs)
|
38 |
print(f"Model outputs: {outputs}")
|
39 |
|
|
|
42 |
probs = logits_per_image.softmax(dim=1)
|
43 |
print(f"Probabilities: {probs}")
|
44 |
|
45 |
+
# Extract probabilities
|
46 |
safe_prob = probs[0][0].item() * 100
|
47 |
unsafe_prob = probs[0][1].item() * 100
|
48 |
+
print(f"Safe: {safe_prob:.2f}%, Unsafe: {unsafe_prob:.2f}%")
|
49 |
|
50 |
# Determine the predicted category
|
51 |
predicted_category = "safe" if safe_prob > unsafe_prob else "unsafe"
|
|
|
|
|
|
|
52 |
return predicted_category, {"safe": f"{safe_prob:.2f}%", "unsafe": f"{unsafe_prob:.2f}%"}
|
53 |
|
54 |
except Exception as e:
|
55 |
print(f"Error during classification: {e}")
|
56 |
return f"Error: {str(e)}", {}
|
57 |
|
58 |
+
# Gradio interface
|
59 |
iface = gr.Interface(
|
60 |
fn=classify_image,
|
61 |
inputs=gr.Image(type="pil"),
|
62 |
outputs=[
|
63 |
+
gr.Textbox(label="Predicted Category"),
|
64 |
+
gr.Label(label="Probabilities"),
|
65 |
],
|
66 |
title="Content Safety Classification",
|
67 |
description="Upload an image to classify it as 'safe' or 'unsafe' with corresponding probabilities.",
|
68 |
)
|
69 |
|
|
|
70 |
if __name__ == "__main__":
|
71 |
print("Launching Gradio interface...")
|
72 |
iface.launch()
|
|
|
91 |
|
92 |
|
93 |
|
94 |
+
|
95 |
|
96 |
|