Spaces:
Sleeping
Sleeping
Commit
·
429fbf1
1
Parent(s):
998f789
latest changes
Browse files
app.py
CHANGED
@@ -7,9 +7,9 @@ import json
|
|
7 |
from PIL import Image
|
8 |
from torchvision import transforms
|
9 |
from huggingface_hub import hf_hub_download
|
10 |
-
import time
|
11 |
|
12 |
-
# --- 1. Define Model Architecture
|
13 |
class SmallVideoClassifier(torch.nn.Module):
|
14 |
def __init__(self, num_classes=2, num_frames=8):
|
15 |
super(SmallVideoClassifier, self).__init__()
|
@@ -58,7 +58,7 @@ CLASS_LABELS = ["Non-violence", "Violence"]
|
|
58 |
if NUM_CLASSES != len(CLASS_LABELS):
|
59 |
print(f"Warning: NUM_CLASSES in config ({NUM_CLASSES}) does not match hardcoded CLASS_LABELS length ({len(CLASS_LABELS)}). Adjust CLASS_LABELS if needed.")
|
60 |
|
61 |
-
device = torch.device("cpu")
|
62 |
print(f"Using device: {device}")
|
63 |
|
64 |
model = SmallVideoClassifier(num_classes=NUM_CLASSES, num_frames=NUM_FRAMES)
|
@@ -78,34 +78,30 @@ transform = transforms.Compose([
|
|
78 |
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
|
79 |
])
|
80 |
|
81 |
-
# ---
|
82 |
-
|
83 |
-
# Initialize global state for the generator function (before the predict function)
|
84 |
-
frame_buffer = [] # Buffer for collecting frames for model input
|
85 |
current_prediction_label = "Initializing..."
|
86 |
-
current_probabilities = {label: 0.0 for label in CLASS_LABELS}
|
87 |
|
|
|
88 |
def predict_live_frames(input_frame):
|
89 |
-
global frame_buffer, current_prediction_label, current_probabilities
|
90 |
|
91 |
if input_frame is None:
|
92 |
-
# If no frame is received (e.g., webcam not active
|
93 |
dummy_frame = np.zeros((200, 400, 3), dtype=np.uint8)
|
94 |
cv2.putText(dummy_frame, "Waiting for webcam input...", (10, 100), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 255), 2)
|
95 |
yield dummy_frame
|
96 |
-
return
|
97 |
|
98 |
-
# Gradio Webcam gives NumPy array (H, W, C) in RGB
|
99 |
pil_image = Image.fromarray(input_frame)
|
100 |
-
|
101 |
-
# Apply transformations (outputs C, H, W tensor)
|
102 |
processed_frame_tensor = transform(pil_image)
|
103 |
frame_buffer.append(processed_frame_tensor)
|
104 |
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
input_tensor = torch.stack(frame_buffer, dim=0).unsqueeze(0).to(device)
|
109 |
|
110 |
with torch.no_grad():
|
111 |
outputs = model(input_tensor)
|
@@ -115,15 +111,8 @@ def predict_live_frames(input_frame):
|
|
115 |
current_prediction_label = f"Class: {CLASS_LABELS[predicted_class_idx]}"
|
116 |
current_probabilities = {CLASS_LABELS[i]: prob.item() for i, prob in enumerate(probabilities[0])}
|
117 |
|
118 |
-
|
119 |
-
|
120 |
-
slide_window_by = 1 # Predict every frame (most "real-time" feel but highest compute)
|
121 |
-
# Or: NUM_FRAMES // 2 (e.g., predict every 4 frames for NUM_FRAMES=8)
|
122 |
-
# Or: NUM_FRAMES (non-overlapping windows, less frequent updates)
|
123 |
-
frame_buffer = frame_buffer[slide_window_by:]
|
124 |
-
|
125 |
-
# --- Draw Prediction on the current input frame ---
|
126 |
-
# Convert the input_frame (RGB NumPy array) to BGR for OpenCV drawing
|
127 |
display_frame = cv2.cvtColor(input_frame, cv2.COLOR_RGB2BGR)
|
128 |
|
129 |
# Draw the main prediction label
|
@@ -132,42 +121,66 @@ def predict_live_frames(input_frame):
|
|
132 |
font_scale = 1.0
|
133 |
font_thickness = 2
|
134 |
|
135 |
-
# Draw outline first for better readability
|
136 |
cv2.putText(display_frame, current_prediction_label, (10, 40),
|
137 |
cv2.FONT_HERSHEY_SIMPLEX, font_scale, text_outline_color, font_thickness + 2, cv2.LINE_AA)
|
138 |
-
# Draw actual text
|
139 |
cv2.putText(display_frame, current_prediction_label, (10, 40),
|
140 |
cv2.FONT_HERSHEY_SIMPLEX, font_scale, text_color, font_thickness, cv2.LINE_AA)
|
141 |
|
142 |
-
# Draw probabilities for all classes
|
143 |
-
y_offset = 80
|
144 |
for label, prob in current_probabilities.items():
|
145 |
prob_text = f"{label}: {prob:.2f}"
|
146 |
cv2.putText(display_frame, prob_text, (10, y_offset),
|
147 |
cv2.FONT_HERSHEY_SIMPLEX, 0.7, text_outline_color, 2, cv2.LINE_AA)
|
148 |
cv2.putText(display_frame, prob_text, (10, y_offset),
|
149 |
-
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 0), 1, cv2.LINE_AA)
|
150 |
-
y_offset += 30
|
151 |
|
152 |
-
# Yield the processed frame back to Gradio for display
|
153 |
-
# Gradio expects RGB NumPy array for video/image components
|
154 |
yield cv2.cvtColor(display_frame, cv2.COLOR_BGR2RGB)
|
155 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
156 |
|
157 |
-
|
158 |
-
iface = gr.Interface(
|
159 |
-
fn=predict_live_frames,
|
160 |
-
# CORRECTED: Use gr.Video with sources=["webcam"] for webcam input
|
161 |
-
inputs=gr.Video(sources=["webcam"], streaming=True, label="Live Webcam Feed for Violence Detection"),
|
162 |
-
# Outputs are updated continuously by the generator
|
163 |
-
outputs=gr.Image(type="numpy", label="Live Prediction Output"), # Using Image as output for continuous frames
|
164 |
-
title="Real-time Violence Detection with SmallVideoClassifier (Webcam)",
|
165 |
-
description=(
|
166 |
-
"This model detects violence in a live webcam feed. "
|
167 |
-
"Predictions (Class and Probabilities) will be displayed on each frame. "
|
168 |
-
"Please allow webcam access when prompted."
|
169 |
-
),
|
170 |
-
allow_flagging="never", # Disable flagging on Hugging Face Spaces
|
171 |
-
)
|
172 |
-
|
173 |
-
iface.launch()
|
|
|
7 |
from PIL import Image
|
8 |
from torchvision import transforms
|
9 |
from huggingface_hub import hf_hub_download
|
10 |
+
import time
|
11 |
|
12 |
+
# --- 1. Define Model Architecture ---
|
13 |
class SmallVideoClassifier(torch.nn.Module):
|
14 |
def __init__(self, num_classes=2, num_frames=8):
|
15 |
super(SmallVideoClassifier, self).__init__()
|
|
|
58 |
if NUM_CLASSES != len(CLASS_LABELS):
|
59 |
print(f"Warning: NUM_CLASSES in config ({NUM_CLASSES}) does not match hardcoded CLASS_LABELS length ({len(CLASS_LABELS)}). Adjust CLASS_LABELS if needed.")
|
60 |
|
61 |
+
device = torch.device("cpu")
|
62 |
print(f"Using device: {device}")
|
63 |
|
64 |
model = SmallVideoClassifier(num_classes=NUM_CLASSES, num_frames=NUM_FRAMES)
|
|
|
78 |
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
|
79 |
])
|
80 |
|
81 |
+
# --- Global state for the generator function ---
|
82 |
+
frame_buffer = []
|
|
|
|
|
83 |
current_prediction_label = "Initializing..."
|
84 |
+
current_probabilities = {label: 0.0 for label in CLASS_LABELS}
|
85 |
|
86 |
+
# --- 4. Gradio Live Inference Function (Generator) ---
|
87 |
def predict_live_frames(input_frame):
|
88 |
+
global frame_buffer, current_prediction_label, current_probabilities
|
89 |
|
90 |
if input_frame is None:
|
91 |
+
# If no frame is received (e.g., webcam not active or disconnected)
|
92 |
dummy_frame = np.zeros((200, 400, 3), dtype=np.uint8)
|
93 |
cv2.putText(dummy_frame, "Waiting for webcam input...", (10, 100), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 255), 2)
|
94 |
yield dummy_frame
|
95 |
+
return
|
96 |
|
|
|
97 |
pil_image = Image.fromarray(input_frame)
|
|
|
|
|
98 |
processed_frame_tensor = transform(pil_image)
|
99 |
frame_buffer.append(processed_frame_tensor)
|
100 |
|
101 |
+
slide_window_by = 1
|
102 |
+
|
103 |
+
if len(frame_buffer) >= NUM_FRAMES:
|
104 |
+
input_tensor = torch.stack(frame_buffer[-NUM_FRAMES:], dim=0).unsqueeze(0).to(device)
|
105 |
|
106 |
with torch.no_grad():
|
107 |
outputs = model(input_tensor)
|
|
|
111 |
current_prediction_label = f"Class: {CLASS_LABELS[predicted_class_idx]}"
|
112 |
current_probabilities = {CLASS_LABELS[i]: prob.item() for i, prob in enumerate(probabilities[0])}
|
113 |
|
114 |
+
frame_buffer = frame_buffer[slide_window_by:]
|
115 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
116 |
display_frame = cv2.cvtColor(input_frame, cv2.COLOR_RGB2BGR)
|
117 |
|
118 |
# Draw the main prediction label
|
|
|
121 |
font_scale = 1.0
|
122 |
font_thickness = 2
|
123 |
|
|
|
124 |
cv2.putText(display_frame, current_prediction_label, (10, 40),
|
125 |
cv2.FONT_HERSHEY_SIMPLEX, font_scale, text_outline_color, font_thickness + 2, cv2.LINE_AA)
|
|
|
126 |
cv2.putText(display_frame, current_prediction_label, (10, 40),
|
127 |
cv2.FONT_HERSHEY_SIMPLEX, font_scale, text_color, font_thickness, cv2.LINE_AA)
|
128 |
|
129 |
+
# Draw probabilities for all classes
|
130 |
+
y_offset = 80
|
131 |
for label, prob in current_probabilities.items():
|
132 |
prob_text = f"{label}: {prob:.2f}"
|
133 |
cv2.putText(display_frame, prob_text, (10, y_offset),
|
134 |
cv2.FONT_HERSHEY_SIMPLEX, 0.7, text_outline_color, 2, cv2.LINE_AA)
|
135 |
cv2.putText(display_frame, prob_text, (10, y_offset),
|
136 |
+
cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 0), 1, cv2.LINE_AA)
|
137 |
+
y_offset += 30
|
138 |
|
|
|
|
|
139 |
yield cv2.cvtColor(display_frame, cv2.COLOR_BGR2RGB)
|
140 |
|
141 |
+
# --- 5. Gradio Blocks Interface Setup ---
|
142 |
+
with gr.Blocks(
|
143 |
+
title="Real-time Violence Detection", # Title for the browser tab
|
144 |
+
theme=gr.themes.Default(primary_hue=gr.Color(c50='#e0f7fa', c100='#b2ebf2', c200='#80deea', c300='#4dd0e1', c400='#26c6da', c500='#00bcd4', c600='#00acc1', c700='#0097a7', c800='#00838f', c900='#006064', ca50='#84ffff', ca100='#18ffff', ca200='#00e5ff', ca400='#00b8d4')) # Optional: A subtle theme change
|
145 |
+
) as demo:
|
146 |
+
# Optional: Display a title and description clearly, even without buttons
|
147 |
+
gr.Markdown(
|
148 |
+
"""
|
149 |
+
# 🎬 Real-time Violence Detection
|
150 |
+
**Live Feed with Constant Predictions**
|
151 |
+
|
152 |
+
This model analyzes your live webcam feed for violence, displaying the predicted class and probabilities on the screen.
|
153 |
+
Please grant webcam access when prompted by your browser.
|
154 |
+
"""
|
155 |
+
)
|
156 |
+
|
157 |
+
with gr.Row():
|
158 |
+
# Input: Live webcam feed
|
159 |
+
# We need to set a minimum height and width to ensure the video feed is displayed reasonably
|
160 |
+
video_input = gr.Video(
|
161 |
+
sources=["webcam"],
|
162 |
+
streaming=True,
|
163 |
+
label="Live Webcam Feed",
|
164 |
+
# Optional: Set dimensions for the video display
|
165 |
+
height=480, # or None for auto
|
166 |
+
width=640 # or None for auto
|
167 |
+
)
|
168 |
+
|
169 |
+
# Output: Image component to display processed frames
|
170 |
+
video_output = gr.Image(
|
171 |
+
type="numpy",
|
172 |
+
label="Processed Feed with Predictions",
|
173 |
+
# Optional: Set dimensions to match input or your preference
|
174 |
+
height=480, # or None for auto
|
175 |
+
width=640 # or None for auto
|
176 |
+
)
|
177 |
+
|
178 |
+
# Connect the video stream directly to the prediction function
|
179 |
+
# The 'stream' event on gr.Video is triggered as new frames arrive from the webcam.
|
180 |
+
video_input.stream(
|
181 |
+
predict_live_frames, # The function to call for each frame
|
182 |
+
inputs=video_input, # Pass the video_input component itself as input
|
183 |
+
outputs=video_output # Update the video_output component
|
184 |
+
)
|
185 |
|
186 |
+
demo.launch()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|