sam2 update

app.py

@@ -3,10 +3,6 @@ VideoMaMa Gradio Demo
 Interactive video matting with SAM2 mask tracking
 """
 
-import sys
-sys.path.append("../")
-sys.path.append("../../")
-
 import os
 import json
 import time
@@ -379,7 +375,8 @@ button {border-radius: 8px !important;}
 """
 
 # Build Gradio interface
-with gr.Blocks(css=custom_css, title="VideoMaMa Demo") as demo:
+with gr.Blocks(title="VideoMaMa Demo") as demo:
+    gr.HTML(f"<style>{custom_css}</style>")
     gr.HTML('<div class="title-text">VideoMaMa Interactive Demo</div>')
     gr.Markdown(
         '<div class="description-text">🎬 Upload a video → 🖱️ Click to mark object → ✅ Generate masks → 🎨 Run VideoMaMa</div>'

sam2.1_hiera_large.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2647878d5dfa5098f2f8649825738a9345572bae2d4350a2468587ece47dd318
+size 898083611

sam2_wrapper.py

@@ -3,9 +3,7 @@ SAM2 Wrapper for Video Mask Tracking
 Handles mask generation and propagation through video
 """
 
-import sys
-sys.path.append("/home/cvlab19/project/samuel/CVPR/sam2")
-
+import os
 import cv2
 import numpy as np
 import torch
@@ -163,8 +161,22 @@ def load_sam2_tracker(device="cuda"):
     Returns:
         SAM2VideoTracker instance
     """
-    checkpoint_path = "/home/cvlab19/project/samuel/CVPR/sam2/checkpoints/sam2.1_hiera_large.pt"
-    config_file = "configs/sam2.1/sam2.1_hiera_l.yaml"
+    # Use relative paths that work on Hugging Face Space
+    # The checkpoint file should be in the root directory or checkpoints/
+    checkpoint_path = "sam2.1_hiera_large.pt"
+    config_file = "sam2_hiera_l.yaml"
+    
+    # Check if checkpoint exists
+    if not os.path.exists(checkpoint_path):
+        # Try alternative path
+        alt_checkpoint_path = os.path.join("checkpoints", "sam2.1_hiera_large.pt")
+        if os.path.exists(alt_checkpoint_path):
+            checkpoint_path = alt_checkpoint_path
+        else:
+            raise FileNotFoundError(
+                f"SAM2 checkpoint not found at {checkpoint_path} or {alt_checkpoint_path}. "
+                "Please run download_checkpoints.sh first or ensure sam2.1_hiera_large.pt is in the root directory."
+            )
     
     print(f"Loading SAM2 from {checkpoint_path}...")
     tracker = SAM2VideoTracker(checkpoint_path, config_file, device)

videomama_wrapper.py

@@ -3,10 +3,7 @@ VideoMaMa Inference Wrapper
 Handles video matting with mask conditioning
 """
 
-import sys
-sys.path.append("../")
-sys.path.append("../../")
-
+import os
 import torch
 import numpy as np
 from PIL import Image
@@ -70,9 +67,23 @@ def load_videomama_pipeline(device="cuda"):
     Returns:
         VideoInferencePipeline instance
     """
-    # Local paths for testing
-    base_model_path = "/home/cvlab19/project/samuel/data/CVPR/pretrained_models/stable-video-diffusion-img2vid-xt"
-    unet_checkpoint_path = "/home/cvlab19/project/samuel/data/CVPR/pretrained_models/videomama"
+    # Use relative paths for Hugging Face Space
+    # Checkpoints should be downloaded via download_checkpoints.sh
+    base_model_path = os.path.join("checkpoints", "stable-video-diffusion-img2vid-xt")
+    unet_checkpoint_path = os.path.join("checkpoints", "videomama")
+    
+    # Check if checkpoints exist
+    if not os.path.exists(base_model_path):
+        raise FileNotFoundError(
+            f"SVD base model not found at {base_model_path}. "
+            "Please run download_checkpoints.sh first."
+        )
+    
+    if not os.path.exists(unet_checkpoint_path):
+        raise FileNotFoundError(
+            f"VideoMaMa checkpoint not found at {unet_checkpoint_path}. "
+            "Please run download_checkpoints.sh first."
+        )
     
     print(f"Loading VideoMaMa pipeline from {unet_checkpoint_path}...")