merge with hf

app.py

@@ -443,7 +443,7 @@ with gr.Blocks(css=css, title="DKT", head=head_html) as demo:
 if __name__ == '__main__':
     
     #* main code, model and moge model initialization
-    demo.queue().launch(share = True)
+    demo.queue().launch()
     
  
 

debug.py

@@ -1,77 +0,0 @@
-import torch
-import os
-from loguru import logger
-from moge.model.v2 import MoGeModel
-
-from tools.eval_utils import  colorize_depth_map
-
-import cv2
-
-video_path = 'examples/1.mp4'
-frames = []
-
-cap = cv2.VideoCapture(video_path)
-while True:
-    ret, frame = cap.read()
-    if not ret:
-        break
-    frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
-    frames.append(frame_rgb)
-cap.release()
-
-print(f"Loaded {len(frames)} frames from {video_path}")
-
-
-
-# device= torch.device("cuda" if torch.cuda.is_available() else "cpu")
-
-# cached_model_path = 'checkpoints/moge_ckpt/moge-2-vitl-normal/model.pt'
-# if os.path.exists(cached_model_path):
-#     logger.info(f"Found cached model at {cached_model_path}, loading from cache...")
-#     moge_pipe = MoGeModel.from_pretrained(cached_model_path).to(device)
-# else:
-#     logger.info(f"Cache not found at {cached_model_path}, downloading from HuggingFace...")
-#     os.makedirs(os.path.dirname(cached_model_path), exist_ok=True)
-#     moge_pipe = MoGeModel.from_pretrained('Ruicheng/moge-2-vitl-normal', cache_dir=os.path.dirname(cached_model_path)).to(device)
-
-
-
-
-#* save small video 
-
-# import numpy as np
-
-# # Save the first 5 frames as new mp4
-# out_path = "first5_output.mp4"
-# frame_subset = frames[:5]
-# target_width, target_height = 832, 480  # width, height as required
-# if len(frame_subset) > 0:
-#     fourcc = cv2.VideoWriter_fourcc(*'mp4v')
-#     out = cv2.VideoWriter(out_path, fourcc, 15, (target_width, target_height))
-#     for f in frame_subset:
-#         if f.dtype != np.uint8:
-#             f = np.clip(f, 0, 255).astype(np.uint8)
-#         # Resize the frame to 832x430 before writing
-#         resized_frame = cv2.resize(f, (target_width, target_height), interpolation=cv2.INTER_AREA)
-#         out.write(resized_frame)
-#     out.release()
-#     print(f"Saved first 5 frames to {out_path} (resized to {target_width}x{target_height})")
-# else:
-#     print("No frames to save.")
-
-# import torch 
-
-# demo_input = frames[0]
-# demo_input = torch.tensor(demo_input / 255, dtype=torch.float32, device=device).permute(2, 0, 1) 
-
-# print(demo_input.max(),demo_input.min(), demo_input.shape)
-# moge_prediction = moge_pipe.infer(demo_input)
-
-
-# moge_depth = moge_prediction['depth'].cpu().numpy()
-# logger.info(f'moge input shape: {demo_input.shape}, input image mean: {demo_input.mean()}, std:{demo_input.std()}, moge_depth:{moge_depth.mean()}, moge_depth: {moge_depth.min()}, moge_depth.max():{moge_depth.max()}')
-# print(f'moge input shape: {demo_input.shape}, input image mean: {demo_input.mean()}, std:{demo_input.std()}, moge_depth:{moge_depth.mean()}, moge_depth: {moge_depth.min()}, moge_depth.max():{moge_depth.max()}')
-
-# moge_depth = colorize_depth_map(moge_depth)
-
-# moge_depth.save('debug.jpg')

debug2.py

@@ -1,11 +0,0 @@
-
-
-from app import * 
-
-video_file = 'examples/1.mp4'
-
-results = process_video(video_file,'1.3B',1,1)
-
-
-for k in results:
-    print(k)