Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,120 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import DetrImageProcessor, DetrForObjectDetection
|
2 |
+
import torch
|
3 |
+
from PIL import Image,ImageDraw
|
4 |
+
import requests
|
5 |
+
import gradio as gr
|
6 |
+
from gtts import gTTS
|
7 |
+
import random
|
8 |
+
from collections import Counter
|
9 |
+
|
10 |
+
|
11 |
+
url = "http://images.cocodataset.org/val2017/000000039769.jpg"
|
12 |
+
image = Image.open(requests.get(url, stream=True).raw)
|
13 |
+
|
14 |
+
# you can specify the revision tag if you don't want the timm dependency
|
15 |
+
processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50", revision="no_timm")
|
16 |
+
model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50", revision="no_timm")
|
17 |
+
|
18 |
+
inputs = processor(images=image, return_tensors="pt")
|
19 |
+
outputs = model(**inputs)
|
20 |
+
|
21 |
+
# convert outputs (bounding boxes and class logits) to COCO API
|
22 |
+
# let's only keep detections with score > 0.9
|
23 |
+
target_sizes = torch.tensor([image.size[::-1]])
|
24 |
+
results = processor.post_process_object_detection(outputs, target_sizes=target_sizes, threshold=0.9)[0]
|
25 |
+
|
26 |
+
for score, label, box in zip(results["scores"], results["labels"], results["boxes"]):
|
27 |
+
box = [round(i, 2) for i in box.tolist()]
|
28 |
+
print(
|
29 |
+
f"Detected {model.config.id2label[label.item()]} with confidence "
|
30 |
+
f"{round(score.item(), 3)} at location {box}"
|
31 |
+
)
|
32 |
+
|
33 |
+
|
34 |
+
# Load model and processor
|
35 |
+
model_name = "facebook/detr-resnet-50"
|
36 |
+
processor = DetrImageProcessor.from_pretrained(model_name)
|
37 |
+
model = DetrForObjectDetection.from_pretrained(model_name)
|
38 |
+
|
39 |
+
# Move model to GPU if available
|
40 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
41 |
+
model.to(device)
|
42 |
+
|
43 |
+
# Function to generate random colors
|
44 |
+
def random_color():
|
45 |
+
return "#{:02x}{:02x}{:02x}".format(random.randint(100, 255), random.randint(100, 255), random.randint(100, 255))
|
46 |
+
|
47 |
+
# Object detection function
|
48 |
+
def detect_objects(image):
|
49 |
+
# Resize image for better detection
|
50 |
+
image = image.resize((800, 800))
|
51 |
+
|
52 |
+
# Process image
|
53 |
+
inputs = processor(images=image, return_tensors="pt").to(device)
|
54 |
+
|
55 |
+
with torch.no_grad():
|
56 |
+
outputs = model(**inputs)
|
57 |
+
|
58 |
+
# Extract bounding boxes and labels
|
59 |
+
target_sizes = [image.size[::-1]]
|
60 |
+
results = processor.post_process_object_detection(outputs, target_sizes=target_sizes)[0]
|
61 |
+
|
62 |
+
# Apply confidence threshold
|
63 |
+
keep = results["scores"] > 0.5
|
64 |
+
boxes = results["boxes"][keep]
|
65 |
+
labels = results["labels"][keep]
|
66 |
+
|
67 |
+
# Create a copy of the image
|
68 |
+
image_draw = image.copy()
|
69 |
+
draw = ImageDraw.Draw(image_draw)
|
70 |
+
|
71 |
+
label_counts = Counter()
|
72 |
+
colors = {}
|
73 |
+
|
74 |
+
# Draw bounding boxes and count labels
|
75 |
+
for box, label in zip(boxes, labels):
|
76 |
+
box = [int(i) for i in box.tolist()]
|
77 |
+
label_text = model.config.id2label[label.item()]
|
78 |
+
|
79 |
+
label_counts[label_text] += 1 # Count occurrences
|
80 |
+
|
81 |
+
if label_text not in colors:
|
82 |
+
colors[label_text] = random_color()
|
83 |
+
|
84 |
+
draw.rectangle(box, outline=colors[label_text], width=5)
|
85 |
+
|
86 |
+
# Prepare HTML output for labels
|
87 |
+
styled_labels = [
|
88 |
+
f"<span style='background-color:{colors[label]}; color:white; padding:8px 15px; border-radius:10px; margin-right:10px;'>"
|
89 |
+
f"{label} (x{count})</span>"
|
90 |
+
for label, count in label_counts.items()
|
91 |
+
]
|
92 |
+
|
93 |
+
labels_html = "<div style='display:flex; flex-wrap:wrap; gap:10px;'>" + " ".join(styled_labels) + "</div>"
|
94 |
+
|
95 |
+
# Convert detected objects into speech
|
96 |
+
detected_objects = ", ".join([f"{label} ({count} times)" for label, count in label_counts.items()])
|
97 |
+
description = f"I detected the following objects: {detected_objects}." if detected_objects else "No objects detected, please try another image."
|
98 |
+
|
99 |
+
# Save audio
|
100 |
+
audio_path = "detected_objects.mp3"
|
101 |
+
tts = gTTS(description)
|
102 |
+
tts.save(audio_path)
|
103 |
+
|
104 |
+
return image_draw, labels_html, audio_path
|
105 |
+
|
106 |
+
# Gradio Interface
|
107 |
+
interface = gr.Interface(
|
108 |
+
fn=detect_objects,
|
109 |
+
inputs=gr.Image(type="pil", label="Upload an Image"),
|
110 |
+
outputs=[
|
111 |
+
gr.Image(label="Detected Objects"),
|
112 |
+
gr.HTML(label="Detected Labels"),
|
113 |
+
gr.Audio(label="Audio Description")
|
114 |
+
],
|
115 |
+
title="AI Assistant for Visually Impaired",
|
116 |
+
description="This app detects objects in an image and provides an audio description."
|
117 |
+
)
|
118 |
+
|
119 |
+
# Launch
|
120 |
+
interface.launch()
|