Spaces:
Running
Running
Rivalcoder
commited on
Commit
·
9a2edf3
1
Parent(s):
d674b3d
Add Files
Browse files- app.py +211 -0
- requirements.txt +8 -0
app.py
ADDED
@@ -0,0 +1,211 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import cv2
|
2 |
+
import torch
|
3 |
+
import numpy as np
|
4 |
+
from PIL import Image
|
5 |
+
import torchvision.transforms as transforms
|
6 |
+
from ultralytics import YOLO
|
7 |
+
import time
|
8 |
+
import os
|
9 |
+
import tempfile
|
10 |
+
from flask import Flask, request, jsonify
|
11 |
+
import gradio as gr
|
12 |
+
|
13 |
+
# Initialize Flask app and Gradio interface
|
14 |
+
app = Flask(__name__)
|
15 |
+
|
16 |
+
# Global variable to store detection history
|
17 |
+
detection_history = []
|
18 |
+
|
19 |
+
# Emotion labels
|
20 |
+
emotions = ['Angry', 'Disgust', 'Fear', 'Happy', 'Sad', 'Surprise', 'Neutral']
|
21 |
+
|
22 |
+
# Load models (cache in Hugging Face Space)
|
23 |
+
def load_models():
|
24 |
+
# Face detection model
|
25 |
+
face_model = YOLO('yolov8n-face.pt')
|
26 |
+
|
27 |
+
# Emotion model (simplified version of your CNN)
|
28 |
+
class EmotionCNN(torch.nn.Module):
|
29 |
+
def __init__(self, num_classes=7):
|
30 |
+
super().__init__()
|
31 |
+
self.features = torch.nn.Sequential(
|
32 |
+
torch.nn.Conv2d(1, 64, 3, padding=1),
|
33 |
+
torch.nn.ReLU(),
|
34 |
+
torch.nn.MaxPool2d(2),
|
35 |
+
torch.nn.Conv2d(64, 128, 3, padding=1),
|
36 |
+
torch.nn.ReLU(),
|
37 |
+
torch.nn.MaxPool2d(2),
|
38 |
+
torch.nn.Conv2d(128, 256, 3, padding=1),
|
39 |
+
torch.nn.ReLU(),
|
40 |
+
torch.nn.MaxPool2d(2)
|
41 |
+
)
|
42 |
+
self.classifier = torch.nn.Sequential(
|
43 |
+
torch.nn.Dropout(0.5),
|
44 |
+
torch.nn.Linear(256*6*6, 1024),
|
45 |
+
torch.nn.ReLU(),
|
46 |
+
torch.nn.Dropout(0.5),
|
47 |
+
torch.nn.Linear(1024, num_classes)
|
48 |
+
)
|
49 |
+
|
50 |
+
def forward(self, x):
|
51 |
+
x = self.features(x)
|
52 |
+
x = torch.flatten(x, 1)
|
53 |
+
x = self.classifier(x)
|
54 |
+
return x
|
55 |
+
|
56 |
+
emotion_model = EmotionCNN()
|
57 |
+
# Load your pretrained weights here
|
58 |
+
# emotion_model.load_state_dict(torch.load('emotion_model.pth'))
|
59 |
+
emotion_model.eval()
|
60 |
+
|
61 |
+
return face_model, emotion_model
|
62 |
+
|
63 |
+
face_model, emotion_model = load_models()
|
64 |
+
|
65 |
+
# Preprocessing function
|
66 |
+
def preprocess_face(face_img):
|
67 |
+
transform = transforms.Compose([
|
68 |
+
transforms.Resize((48, 48)),
|
69 |
+
transforms.Grayscale(),
|
70 |
+
transforms.ToTensor(),
|
71 |
+
transforms.Normalize(mean=[0.5], std=[0.5])
|
72 |
+
])
|
73 |
+
face_pil = Image.fromarray(cv2.cvtColor(face_img, cv2.COLOR_BGR2RGB))
|
74 |
+
return transform(face_pil).unsqueeze(0)
|
75 |
+
|
76 |
+
# Process video function
|
77 |
+
def process_video(video_path):
|
78 |
+
global detection_history
|
79 |
+
detection_history = []
|
80 |
+
|
81 |
+
cap = cv2.VideoCapture(video_path)
|
82 |
+
if not cap.isOpened():
|
83 |
+
return {"error": "Could not open video"}
|
84 |
+
|
85 |
+
frame_count = 0
|
86 |
+
fps = cap.get(cv2.CAP_PROP_FPS)
|
87 |
+
frame_skip = int(fps / 3) # Process ~3 frames per second
|
88 |
+
|
89 |
+
while True:
|
90 |
+
ret, frame = cap.read()
|
91 |
+
if not ret:
|
92 |
+
break
|
93 |
+
|
94 |
+
frame_count += 1
|
95 |
+
if frame_count % frame_skip != 0:
|
96 |
+
continue
|
97 |
+
|
98 |
+
# Face detection
|
99 |
+
results = face_model(frame)
|
100 |
+
|
101 |
+
for result in results:
|
102 |
+
boxes = result.boxes
|
103 |
+
if len(boxes) == 0:
|
104 |
+
continue
|
105 |
+
|
106 |
+
for box in boxes:
|
107 |
+
x1, y1, x2, y2 = map(int, box.xyxy[0].tolist())
|
108 |
+
face_img = frame[y1:y2, x1:x2]
|
109 |
+
|
110 |
+
if face_img.size == 0:
|
111 |
+
continue
|
112 |
+
|
113 |
+
# Emotion prediction
|
114 |
+
face_tensor = preprocess_face(face_img)
|
115 |
+
with torch.no_grad():
|
116 |
+
output = emotion_model(face_tensor)
|
117 |
+
prob = torch.nn.functional.softmax(output, dim=1)[0]
|
118 |
+
pred_idx = torch.argmax(output).item()
|
119 |
+
confidence = prob[pred_idx].item()
|
120 |
+
|
121 |
+
detection_history.append({
|
122 |
+
"frame": frame_count,
|
123 |
+
"time": frame_count / fps,
|
124 |
+
"emotion": emotions[pred_idx],
|
125 |
+
"confidence": confidence,
|
126 |
+
"box": [x1, y1, x2, y2]
|
127 |
+
})
|
128 |
+
|
129 |
+
cap.release()
|
130 |
+
|
131 |
+
if not detection_history:
|
132 |
+
return {"error": "No faces detected"}
|
133 |
+
|
134 |
+
return {
|
135 |
+
"detections": detection_history,
|
136 |
+
"summary": {
|
137 |
+
"total_frames": frame_count,
|
138 |
+
"fps": fps,
|
139 |
+
"duration": frame_count / fps
|
140 |
+
}
|
141 |
+
}
|
142 |
+
|
143 |
+
# Flask API endpoint
|
144 |
+
@app.route('/api/predict', methods=['POST'])
|
145 |
+
def api_predict():
|
146 |
+
if 'file' not in request.files:
|
147 |
+
return jsonify({"error": "No file provided"}), 400
|
148 |
+
|
149 |
+
file = request.files['file']
|
150 |
+
if file.filename == '':
|
151 |
+
return jsonify({"error": "No selected file"}), 400
|
152 |
+
|
153 |
+
# Save to temp file
|
154 |
+
temp_path = os.path.join(tempfile.gettempdir(), file.filename)
|
155 |
+
file.save(temp_path)
|
156 |
+
|
157 |
+
# Process video
|
158 |
+
result = process_video(temp_path)
|
159 |
+
|
160 |
+
# Clean up
|
161 |
+
os.remove(temp_path)
|
162 |
+
|
163 |
+
return jsonify(result)
|
164 |
+
|
165 |
+
# Gradio interface
|
166 |
+
def gradio_predict(video):
|
167 |
+
temp_path = os.path.join(tempfile.gettempdir(), video.name)
|
168 |
+
with open(temp_path, 'wb') as f:
|
169 |
+
f.write(video.read())
|
170 |
+
|
171 |
+
result = process_video(temp_path)
|
172 |
+
os.remove(temp_path)
|
173 |
+
|
174 |
+
if "error" in result:
|
175 |
+
return result["error"]
|
176 |
+
|
177 |
+
# Create visualization
|
178 |
+
cap = cv2.VideoCapture(video.name)
|
179 |
+
ret, frame = cap.read()
|
180 |
+
cap.release()
|
181 |
+
|
182 |
+
if ret:
|
183 |
+
# Draw last detection on frame
|
184 |
+
last_det = result["detections"][-1]
|
185 |
+
x1, y1, x2, y2 = last_det["box"]
|
186 |
+
cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 255, 0), 2)
|
187 |
+
cv2.putText(frame, f"{last_det['emotion']} ({last_det['confidence']:.2f})",
|
188 |
+
(x1, y1-10), cv2.FONT_HERSHEY_SIMPLEX, 0.9, (0, 255, 0), 2)
|
189 |
+
|
190 |
+
# Convert to RGB for Gradio
|
191 |
+
frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
|
192 |
+
return frame, result
|
193 |
+
return result
|
194 |
+
|
195 |
+
# Create Gradio interface
|
196 |
+
demo = gr.Interface(
|
197 |
+
fn=gradio_predict,
|
198 |
+
inputs=gr.Video(label="Upload Video"),
|
199 |
+
outputs=[
|
200 |
+
gr.Image(label="Detection Preview"),
|
201 |
+
gr.JSON(label="Results")
|
202 |
+
],
|
203 |
+
title="Video Emotion Detection",
|
204 |
+
description="Upload a video to detect emotions in faces"
|
205 |
+
)
|
206 |
+
|
207 |
+
# Mount Gradio app
|
208 |
+
app = gr.mount_gradio_app(app, demo, path="/")
|
209 |
+
|
210 |
+
if __name__ == "__main__":
|
211 |
+
app.run(host="0.0.0.0", port=7860)
|
requirements.txt
ADDED
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
torch
|
2 |
+
torchvision
|
3 |
+
opencv-python
|
4 |
+
ultralytics
|
5 |
+
gradio
|
6 |
+
flask
|
7 |
+
numpy
|
8 |
+
Pillow
|