Update app.py
Browse files
app.py
CHANGED
|
@@ -4,17 +4,8 @@
|
|
| 4 |
import torch
|
| 5 |
import tempfile
|
| 6 |
import os
|
| 7 |
-
import logging
|
| 8 |
-
from pathlib import Path
|
| 9 |
import time
|
| 10 |
-
|
| 11 |
-
# Configure logging
|
| 12 |
-
logging.basicConfig(
|
| 13 |
-
level=logging.INFO,
|
| 14 |
-
format='[%(asctime)s] %(levelname)s: %(message)s',
|
| 15 |
-
datefmt='%H:%M:%S'
|
| 16 |
-
)
|
| 17 |
-
logger = logging.getLogger(__name__)
|
| 18 |
|
| 19 |
# Import model loaders
|
| 20 |
from model_loaders import load_sam2, load_matanyone, pose
|
|
@@ -25,7 +16,6 @@ def log_and_progress(progress_callback, stage, progress, message):
|
|
| 25 |
"""Unified logging and progress reporting"""
|
| 26 |
timestamp = time.strftime("%H:%M:%S")
|
| 27 |
log_msg = f"[{timestamp}] Stage {stage} ({progress:.0%}): {message}"
|
| 28 |
-
logger.info(log_msg)
|
| 29 |
print(log_msg, flush=True)
|
| 30 |
progress_callback(stage, progress, message)
|
| 31 |
|
|
@@ -35,18 +25,18 @@ def process_video(input_path, output_path, progress_callback):
|
|
| 35 |
# ============================================================
|
| 36 |
# STAGE 1: SEGMENTATION (Load models, read video, segment all frames)
|
| 37 |
# ============================================================
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
stage1_start = time.time()
|
| 42 |
|
| 43 |
# 1.1: Load SAM2
|
| 44 |
log_and_progress(progress_callback, 1, 0.0, "Loading SAM2 model...")
|
| 45 |
try:
|
| 46 |
sam_predictor = load_sam2()
|
| 47 |
-
|
| 48 |
except Exception as e:
|
| 49 |
-
|
| 50 |
raise
|
| 51 |
|
| 52 |
# 1.2: Load video
|
|
@@ -61,9 +51,9 @@ def process_video(input_path, output_path, progress_callback):
|
|
| 61 |
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
| 62 |
total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
|
| 63 |
|
| 64 |
-
|
| 65 |
except Exception as e:
|
| 66 |
-
|
| 67 |
raise
|
| 68 |
|
| 69 |
# 1.3: Read all frames
|
|
@@ -75,7 +65,7 @@ def process_video(input_path, output_path, progress_callback):
|
|
| 75 |
break
|
| 76 |
frames.append(frame)
|
| 77 |
cap.release()
|
| 78 |
-
|
| 79 |
|
| 80 |
# 1.4: Segment all frames
|
| 81 |
log_and_progress(progress_callback, 1, 0.2, "Starting person segmentation...")
|
|
@@ -117,31 +107,31 @@ def process_video(input_path, output_path, progress_callback):
|
|
| 117 |
masks.append(np.zeros((h, w), dtype=bool))
|
| 118 |
|
| 119 |
except Exception as e:
|
| 120 |
-
|
| 121 |
h, w = frame.shape[:2]
|
| 122 |
masks.append(np.zeros((h, w), dtype=bool))
|
| 123 |
|
| 124 |
stage1_time = time.time() - stage1_start
|
| 125 |
-
|
| 126 |
-
|
| 127 |
-
|
| 128 |
-
|
| 129 |
|
| 130 |
# ============================================================
|
| 131 |
# STAGE 2: MATTING (Refine all masks, smooth, write video)
|
| 132 |
# ============================================================
|
| 133 |
-
|
| 134 |
-
|
| 135 |
-
|
| 136 |
stage2_start = time.time()
|
| 137 |
|
| 138 |
# 2.1: Load MatAnyone
|
| 139 |
log_and_progress(progress_callback, 2, 0.0, "Loading MatAnyone model...")
|
| 140 |
try:
|
| 141 |
matanyone = load_matanyone()
|
| 142 |
-
|
| 143 |
except Exception as e:
|
| 144 |
-
|
| 145 |
raise
|
| 146 |
|
| 147 |
# 2.2: Process all frames with MatAnyone
|
|
@@ -168,10 +158,10 @@ def process_video(input_path, output_path, progress_callback):
|
|
| 168 |
alphas.append(alpha)
|
| 169 |
|
| 170 |
except Exception as e:
|
| 171 |
-
|
| 172 |
alphas.append(np.zeros((frame.shape[0], frame.shape[1]), dtype=np.float32))
|
| 173 |
|
| 174 |
-
|
| 175 |
|
| 176 |
# 2.3: Temporal smoothing
|
| 177 |
log_and_progress(progress_callback, 2, 0.65, "Applying temporal smoothing to eliminate jitter...")
|
|
@@ -181,6 +171,9 @@ def process_video(input_path, output_path, progress_callback):
|
|
| 181 |
half_window = window_size // 2
|
| 182 |
|
| 183 |
for i in range(len(alphas)):
|
|
|
|
|
|
|
|
|
|
| 184 |
start_idx = max(0, i - half_window)
|
| 185 |
end_idx = min(len(alphas), i + half_window + 1)
|
| 186 |
window_alphas = alphas[start_idx:end_idx]
|
|
@@ -189,11 +182,11 @@ def process_video(input_path, output_path, progress_callback):
|
|
| 189 |
smoothed = np.mean(window_alphas, axis=0)
|
| 190 |
smoothed_alphas.append(smoothed)
|
| 191 |
|
| 192 |
-
|
| 193 |
alphas = smoothed_alphas
|
| 194 |
|
| 195 |
except Exception as e:
|
| 196 |
-
|
| 197 |
|
| 198 |
# 2.4: Write output video
|
| 199 |
log_and_progress(progress_callback, 2, 0.75, "Writing output video...")
|
|
@@ -215,23 +208,23 @@ def process_video(input_path, output_path, progress_callback):
|
|
| 215 |
out.write(output)
|
| 216 |
|
| 217 |
out.release()
|
| 218 |
-
|
| 219 |
|
| 220 |
except Exception as e:
|
| 221 |
-
|
| 222 |
raise
|
| 223 |
|
| 224 |
stage2_time = time.time() - stage2_start
|
| 225 |
total_time = stage1_time + stage2_time
|
| 226 |
|
| 227 |
-
|
| 228 |
-
|
| 229 |
-
|
| 230 |
-
|
| 231 |
-
|
| 232 |
-
|
| 233 |
-
|
| 234 |
-
|
| 235 |
|
| 236 |
log_and_progress(progress_callback, 2, 1.0, "Processing complete!")
|
| 237 |
return output_path
|
|
@@ -251,6 +244,10 @@ def main():
|
|
| 251 |
output_path = tempfile.mktemp(suffix='_output.mp4')
|
| 252 |
|
| 253 |
if st.button("π Process Video", type="primary"):
|
|
|
|
|
|
|
|
|
|
|
|
|
| 254 |
# Progress tracking
|
| 255 |
stage1_progress = st.progress(0, text="Stage 1: Initializing...")
|
| 256 |
stage1_status = st.empty()
|
|
@@ -267,7 +264,6 @@ def update_progress(stage, progress, message):
|
|
| 267 |
stage2_status.info(f"π {message}")
|
| 268 |
|
| 269 |
try:
|
| 270 |
-
logger.info("π¬ Starting video processing...")
|
| 271 |
result_path = process_video(input_path, output_path, update_progress)
|
| 272 |
|
| 273 |
stage1_status.success("β
Stage 1: Segmentation complete!")
|
|
@@ -288,9 +284,15 @@ def update_progress(stage, progress, message):
|
|
| 288 |
st.video(result_path)
|
| 289 |
|
| 290 |
except Exception as e:
|
| 291 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 292 |
st.error(f"β Processing failed: {str(e)}")
|
| 293 |
-
st.error("Check the logs
|
| 294 |
|
| 295 |
finally:
|
| 296 |
# Cleanup
|
|
@@ -302,4 +304,4 @@ def update_progress(stage, progress, message):
|
|
| 302 |
pass
|
| 303 |
|
| 304 |
if __name__ == "__main__":
|
| 305 |
-
main()
|
|
|
|
| 4 |
import torch
|
| 5 |
import tempfile
|
| 6 |
import os
|
|
|
|
|
|
|
| 7 |
import time
|
| 8 |
+
from pathlib import Path
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
|
| 10 |
# Import model loaders
|
| 11 |
from model_loaders import load_sam2, load_matanyone, pose
|
|
|
|
| 16 |
"""Unified logging and progress reporting"""
|
| 17 |
timestamp = time.strftime("%H:%M:%S")
|
| 18 |
log_msg = f"[{timestamp}] Stage {stage} ({progress:.0%}): {message}"
|
|
|
|
| 19 |
print(log_msg, flush=True)
|
| 20 |
progress_callback(stage, progress, message)
|
| 21 |
|
|
|
|
| 25 |
# ============================================================
|
| 26 |
# STAGE 1: SEGMENTATION (Load models, read video, segment all frames)
|
| 27 |
# ============================================================
|
| 28 |
+
print("="*60, flush=True)
|
| 29 |
+
print("STAGE 1: PERSON SEGMENTATION", flush=True)
|
| 30 |
+
print("="*60, flush=True)
|
| 31 |
stage1_start = time.time()
|
| 32 |
|
| 33 |
# 1.1: Load SAM2
|
| 34 |
log_and_progress(progress_callback, 1, 0.0, "Loading SAM2 model...")
|
| 35 |
try:
|
| 36 |
sam_predictor = load_sam2()
|
| 37 |
+
print(f"β
SAM2 loaded successfully", flush=True)
|
| 38 |
except Exception as e:
|
| 39 |
+
print(f"β SAM2 loading failed: {e}", flush=True)
|
| 40 |
raise
|
| 41 |
|
| 42 |
# 1.2: Load video
|
|
|
|
| 51 |
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
| 52 |
total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
|
| 53 |
|
| 54 |
+
print(f"πΉ Video: {width}x{height} @ {fps:.2f}fps, {total_frames} frames", flush=True)
|
| 55 |
except Exception as e:
|
| 56 |
+
print(f"β Video opening failed: {e}", flush=True)
|
| 57 |
raise
|
| 58 |
|
| 59 |
# 1.3: Read all frames
|
|
|
|
| 65 |
break
|
| 66 |
frames.append(frame)
|
| 67 |
cap.release()
|
| 68 |
+
print(f"β
Read {len(frames)} frames", flush=True)
|
| 69 |
|
| 70 |
# 1.4: Segment all frames
|
| 71 |
log_and_progress(progress_callback, 1, 0.2, "Starting person segmentation...")
|
|
|
|
| 107 |
masks.append(np.zeros((h, w), dtype=bool))
|
| 108 |
|
| 109 |
except Exception as e:
|
| 110 |
+
print(f"β Frame {i+1} segmentation failed: {e}", flush=True)
|
| 111 |
h, w = frame.shape[:2]
|
| 112 |
masks.append(np.zeros((h, w), dtype=bool))
|
| 113 |
|
| 114 |
stage1_time = time.time() - stage1_start
|
| 115 |
+
print("="*60, flush=True)
|
| 116 |
+
print(f"β
STAGE 1 COMPLETE in {stage1_time:.1f}s", flush=True)
|
| 117 |
+
print(f" Segmented {len(masks)} frames", flush=True)
|
| 118 |
+
print("="*60, flush=True)
|
| 119 |
|
| 120 |
# ============================================================
|
| 121 |
# STAGE 2: MATTING (Refine all masks, smooth, write video)
|
| 122 |
# ============================================================
|
| 123 |
+
print("="*60, flush=True)
|
| 124 |
+
print("STAGE 2: HIGH-QUALITY MATTING", flush=True)
|
| 125 |
+
print("="*60, flush=True)
|
| 126 |
stage2_start = time.time()
|
| 127 |
|
| 128 |
# 2.1: Load MatAnyone
|
| 129 |
log_and_progress(progress_callback, 2, 0.0, "Loading MatAnyone model...")
|
| 130 |
try:
|
| 131 |
matanyone = load_matanyone()
|
| 132 |
+
print(f"β
MatAnyone loaded successfully", flush=True)
|
| 133 |
except Exception as e:
|
| 134 |
+
print(f"β MatAnyone loading failed: {e}", flush=True)
|
| 135 |
raise
|
| 136 |
|
| 137 |
# 2.2: Process all frames with MatAnyone
|
|
|
|
| 158 |
alphas.append(alpha)
|
| 159 |
|
| 160 |
except Exception as e:
|
| 161 |
+
print(f"β Frame {i+1} matting failed: {e}", flush=True)
|
| 162 |
alphas.append(np.zeros((frame.shape[0], frame.shape[1]), dtype=np.float32))
|
| 163 |
|
| 164 |
+
print(f"β
Matted {len(alphas)} frames", flush=True)
|
| 165 |
|
| 166 |
# 2.3: Temporal smoothing
|
| 167 |
log_and_progress(progress_callback, 2, 0.65, "Applying temporal smoothing to eliminate jitter...")
|
|
|
|
| 171 |
half_window = window_size // 2
|
| 172 |
|
| 173 |
for i in range(len(alphas)):
|
| 174 |
+
if i % 100 == 0:
|
| 175 |
+
print(f"Smoothing frame {i}/{len(alphas)}...", flush=True)
|
| 176 |
+
|
| 177 |
start_idx = max(0, i - half_window)
|
| 178 |
end_idx = min(len(alphas), i + half_window + 1)
|
| 179 |
window_alphas = alphas[start_idx:end_idx]
|
|
|
|
| 182 |
smoothed = np.mean(window_alphas, axis=0)
|
| 183 |
smoothed_alphas.append(smoothed)
|
| 184 |
|
| 185 |
+
print(f"β
Applied {window_size}-frame temporal smoothing", flush=True)
|
| 186 |
alphas = smoothed_alphas
|
| 187 |
|
| 188 |
except Exception as e:
|
| 189 |
+
print(f"β οΈ Smoothing failed: {e}, using unsmoothed alphas", flush=True)
|
| 190 |
|
| 191 |
# 2.4: Write output video
|
| 192 |
log_and_progress(progress_callback, 2, 0.75, "Writing output video...")
|
|
|
|
| 208 |
out.write(output)
|
| 209 |
|
| 210 |
out.release()
|
| 211 |
+
print(f"β
Video written to {output_path}", flush=True)
|
| 212 |
|
| 213 |
except Exception as e:
|
| 214 |
+
print(f"β Video writing failed: {e}", flush=True)
|
| 215 |
raise
|
| 216 |
|
| 217 |
stage2_time = time.time() - stage2_start
|
| 218 |
total_time = stage1_time + stage2_time
|
| 219 |
|
| 220 |
+
print("="*60, flush=True)
|
| 221 |
+
print(f"β
STAGE 2 COMPLETE in {stage2_time:.1f}s", flush=True)
|
| 222 |
+
print("="*60, flush=True)
|
| 223 |
+
print(f"π TOTAL PROCESSING TIME: {total_time:.1f}s", flush=True)
|
| 224 |
+
print(f" Stage 1 (Segmentation): {stage1_time:.1f}s", flush=True)
|
| 225 |
+
print(f" Stage 2 (Matting): {stage2_time:.1f}s", flush=True)
|
| 226 |
+
print(f" Average: {total_time/len(frames):.2f}s per frame", flush=True)
|
| 227 |
+
print("="*60, flush=True)
|
| 228 |
|
| 229 |
log_and_progress(progress_callback, 2, 1.0, "Processing complete!")
|
| 230 |
return output_path
|
|
|
|
| 244 |
output_path = tempfile.mktemp(suffix='_output.mp4')
|
| 245 |
|
| 246 |
if st.button("π Process Video", type="primary"):
|
| 247 |
+
print(f"\n\n{'='*60}", flush=True)
|
| 248 |
+
print(f"π¬ NEW VIDEO PROCESSING STARTED", flush=True)
|
| 249 |
+
print(f"{'='*60}\n", flush=True)
|
| 250 |
+
|
| 251 |
# Progress tracking
|
| 252 |
stage1_progress = st.progress(0, text="Stage 1: Initializing...")
|
| 253 |
stage1_status = st.empty()
|
|
|
|
| 264 |
stage2_status.info(f"π {message}")
|
| 265 |
|
| 266 |
try:
|
|
|
|
| 267 |
result_path = process_video(input_path, output_path, update_progress)
|
| 268 |
|
| 269 |
stage1_status.success("β
Stage 1: Segmentation complete!")
|
|
|
|
| 284 |
st.video(result_path)
|
| 285 |
|
| 286 |
except Exception as e:
|
| 287 |
+
print(f"\n{'='*60}", flush=True)
|
| 288 |
+
print(f"βββ PROCESSING FAILED βββ", flush=True)
|
| 289 |
+
print(f"Error: {str(e)}", flush=True)
|
| 290 |
+
print(f"{'='*60}\n", flush=True)
|
| 291 |
+
import traceback
|
| 292 |
+
traceback.print_exc()
|
| 293 |
+
|
| 294 |
st.error(f"β Processing failed: {str(e)}")
|
| 295 |
+
st.error("Check the container logs for full details")
|
| 296 |
|
| 297 |
finally:
|
| 298 |
# Cleanup
|
|
|
|
| 304 |
pass
|
| 305 |
|
| 306 |
if __name__ == "__main__":
|
| 307 |
+
main()
|