refactoring

app.py

@@ -5,7 +5,6 @@
 # LICENSE file in the root directory of this source tree.
 # conda activate hf3.10
 
-import base64
 import gc
 import os
 import shutil
@@ -21,22 +20,24 @@ import torch
 
 sys.path.append("mapanything/")
 
-from hf_utils.css_and_html import (
-    GRADIO_CSS,
-    MEASURE_INSTRUCTIONS_HTML,
+from mapanything.utils.geometry import depthmap_to_world_frame, points_to_normals
+from mapanything.utils.hf_utils.css_and_html import (
     get_acknowledgements_html,
     get_description_html,
     get_gradio_theme,
     get_header_html,
+    GRADIO_CSS,
+    MEASURE_INSTRUCTIONS_HTML,
 )
-from hf_utils.visual_util import predictions_to_glb
-from mapanything.models import MapAnything
-from mapanything.utils.geometry import depthmap_to_world_frame, points_to_normals
+from mapanything.utils.hf_utils.hf_helpers import initialize_mapanything_model
+from mapanything.utils.hf_utils.visual_util import predictions_to_glb
 from mapanything.utils.image import load_images, rgb
 
 
 def get_logo_base64():
     """Convert WAI logo to base64 for embedding in HTML"""
+    import base64
+    
     logo_path = "examples/wai_logo/wai_logo.png"
     try:
         with open(logo_path, "rb") as img_file:
@@ -47,72 +48,39 @@ def get_logo_base64():
         return None
 
 
-print("Initializing and loading MapAnything model...")
-
-
-def load_hf_token():
-    """Load HuggingFace access token from local file"""
-    token_file_paths = [
-        "/home/aknapitsch/hf_token.txt",
-    ]
-
-    for token_path in token_file_paths:
-        if os.path.exists(token_path):
-            try:
-                with open(token_path, "r") as f:
-                    token = f.read().strip()
-                print(f"Loaded HuggingFace token from: {token_path}")
-                return token
-            except Exception as e:
-                print(f"Error reading token from {token_path}: {e}")
-                continue
-        else:
-            print(token_path, "token_path doesnt exist")
-
-    # Also try environment variable
-    # see https://huggingface.co/docs/hub/spaces-overview#managing-secrets on options
-    token = (
-        os.getenv("HF_TOKEN")
-        or os.getenv("HUGGING_FACE_HUB_TOKEN")
-        or os.getenv("HUGGING_FACE_MODEL_TOKEN")
-    )
-    if token:
-        print("Loaded HuggingFace token from environment variable")
-        return token
-
-    print(
-        "Warning: No HuggingFace token found. Model loading may fail for private repositories."
-    )
-    return None
-
-
-def init_hydra_config(config_path, overrides=None):
-    "Initialize Hydra config"
-    import hydra
-
-    config_dir = os.path.dirname(config_path)
-    config_name = os.path.basename(config_path).split(".")[0]
-    relative_path = os.path.relpath(config_dir, os.path.dirname(__file__))
-    hydra.core.global_hydra.GlobalHydra.instance().clear()
-    hydra.initialize(version_base=None, config_path=relative_path)
-    if overrides is not None:
-        cfg = hydra.compose(config_name=config_name, overrides=overrides)
-    else:
-        cfg = hydra.compose(config_name=config_name)
-    return cfg
-
+# MapAnything Configuration
+# high_level_config = {
+#     "path": "configs/train.yaml",
+#     "hf_model_name": "facebook/MapAnything",
+#     "model_str": "mapanything",
+#     "config_overrides": [
+#         "machine=aws",
+#         "model=mapanything",
+#         "model/task=images_only",
+#         "model.encoder.uses_torch_hub=false",
+#     ],
+#     "checkpoint_name": "mapa_curri_24v_13d_48ipg_64g.pth",
+#     "config_name": "config.json",
+#     "trained_with_amp": True,
+#     "trained_with_amp_dtype": "fp16",
+#     "data_norm_type": "dinov2",
+#     "patch_size": 14,
+#     "resolution": 518,
+# }
 
 # MapAnything Configuration
 high_level_config = {
     "path": "configs/train.yaml",
     "hf_model_name": "facebook/map-anything-apache",
+    "model_str": "mapanything",
     "config_overrides": [
         "machine=aws",
         "model=mapanything",
         "model/task=images_only",
         "model.encoder.uses_torch_hub=false",
     ],
-    "checkpoint_path": "https://huggingface.co/facebook/MapAnything/resolve/main/mapa_curri_24v_13d_48ipg_64g.pth",
+    "checkpoint_name": "model.safetensors",
+    "config_name": "config.json",
     "trained_with_amp": True,
     "trained_with_amp_dtype": "fp16",
     "data_norm_type": "dinov2",
@@ -128,7 +96,7 @@ model = None
 # 1) Core model inference
 # -------------------------------------------------------------------------
 @spaces.GPU(duration=120)
-def run_model(target_dir, model_placeholder, apply_mask=True, mask_edges=True):
+def run_model(target_dir, apply_mask=True, mask_edges=True):
     """
     Run the MapAnything model on images in the 'target_dir/images' folder and return predictions.
     """
@@ -143,87 +111,7 @@ def run_model(target_dir, model_placeholder, apply_mask=True, mask_edges=True):
 
     # Initialize model if not already done
     if model is None:
-        print("Initializing MapAnything model...")
-
-        # Initialize Hydra config and create model from configuration
-        cfg = init_hydra_config(
-            high_level_config["path"], overrides=high_level_config["config_overrides"]
-        )
-
-        # Try using from_pretrained first
-        try:
-            print("Loading MapAnything model from_pretrained...")
-            model = MapAnything.from_pretrained(high_level_config["hf_model_name"]).to(
-                device
-            )
-            print("Loading MapAnything model from_pretrained succeeded...")
-        except Exception as e:
-            print(f"from_pretrained failed: {e}")
-            print("Falling back to local configuration approach...")
-            
-            # Create model from local configuration instead of using from_pretrained
-            from mapanything.models import init_model
-
-            model = init_model(
-                model_str=cfg.model.model_str,
-                model_config=cfg.model.model_config,
-                torch_hub_force_reload=high_level_config.get(
-                    "torch_hub_force_reload", False
-                ),
-            )
-
-            # Load the pretrained weights from HuggingFace Hub
-            try:
-                from huggingface_hub import hf_hub_download, list_repo_files
-
-                # First, let's see what files are available in the repository
-                try:
-                    repo_files = list_repo_files(
-                        repo_id=high_level_config["hf_model_name"], token=load_hf_token()
-                    )
-                    print(f"Available files in repository: {repo_files}")
-
-                    checkpoint_filename = "model.safetensors"
-
-                    # Download the model weights
-                    checkpoint_path = hf_hub_download(
-                        repo_id=high_level_config["hf_model_name"],
-                        filename=checkpoint_filename,
-                        token=load_hf_token(),
-                    )
-
-                    # Load the weights
-                    print("start loading checkpoint")
-                    if checkpoint_filename.endswith(".safetensors"):
-                        from safetensors.torch import load_file
-
-                        checkpoint = load_file(checkpoint_path)
-                    else:
-                        checkpoint = torch.load(
-                            checkpoint_path, map_location="cpu", weights_only=True
-                        )
-
-                    print("start loading state_dict")
-                    if "model" in checkpoint:
-                        model.load_state_dict(checkpoint["model"])
-                    elif "state_dict" in checkpoint:
-                        model.load_state_dict(checkpoint["state_dict"])
-                    else:
-                        model.load_state_dict(checkpoint)
-
-                    print(
-                        f"Successfully loaded pretrained weights from HuggingFace Hub ({checkpoint_filename})"
-                    )
-
-                except Exception as inner_e:
-                    print(f"Error listing repository files or loading weights: {inner_e}")
-                    raise inner_e
-
-            except Exception as e:
-                print(f"Warning: Could not load pretrained weights: {e}")
-                print("Proceeding with randomly initialized model...")
-
-        model = model.to(device)
+        model = initialize_mapanything_model(high_level_config, device)
 
     else:
         model = model.to(device)
@@ -400,7 +288,7 @@ def update_measure_view(processed_data, view_index):
 
         if invalid_mask.any():
             # Create a light grey overlay (RGB: 192, 192, 192)
-            overlay_color = np.array([192, 192, 192], dtype=np.uint8)
+            overlay_color = np.array([255, 220, 220], dtype=np.uint8)
 
             # Apply overlay with some transparency
             alpha = 0.5  # Transparency level
@@ -619,7 +507,7 @@ def gradio_demo(
     print("Running MapAnything model...")
     with torch.no_grad():
         predictions, processed_data = run_model(
-            target_dir, None, apply_mask, mask_edges
+            target_dir, apply_mask, mask_edges
         )
 
     # Save predictions
@@ -768,18 +656,9 @@ def process_predictions_for_visualization(predictions, views, high_level_config)
 
     # Check if confidence data is available in any view
     has_confidence_data = False
-    # for view_idx, view in enumerate(views):
-    #     view_key = f"pred{view_idx + 1}"
-    #     if view_key in pred_result and "conf" in pred_result[view_key]:
-    #         has_confidence_data = True
-    #         break
 
     # Process each view
     for view_idx, view in enumerate(views):
-        # view_key = f"pred{view_idx + 1}"
-        # if view_key not in pred_result:
-        #     continue
-
         # Get image
         image = rgb(view["img"], norm_type=high_level_config["data_norm_type"])
         # image = rgb(view["img"], norm_type=high_level_config["data_norm_type"])
@@ -1241,7 +1120,7 @@ with gr.Blocks(theme=theme, css=GRADIO_CSS) as demo:
                             sources=[],
                         )
                         gr.Markdown(
-                            "**Note:** Gray areas indicate regions with no depth information where measurements cannot be taken."
+                            "**Note:** Light-grey areas indicate regions with no depth information where measurements cannot be taken."
                         )
                         measure_text = gr.Markdown("")
 

{hf_utils → mapanything/utils/hf_utils}/css_and_html.py

@@ -164,8 +164,8 @@ def get_acknowledgements_html():
         <h3>Acknowledgements</h3>
         <p>This site builds upon code from:</p>
         <ul style="list-style: none; padding: 0;">
-            <li>🔗 <a href="https://huggingface.co/spaces/Ruicheng/MoGe-2">MoGe-2 on Hugging Face</a></li>
-            <li>🔗 <a href="https://huggingface.co/spaces/facebook/vggt">VGGT on Hugging Face</a></li>
+            <li>🔗 <a href="https://github.com/microsoft/MoGe">MoGe (and MoGe2) on GitHub (and HuggingFace)</a></li>
+            <li>🔗 <a href="https://github.com/facebookresearch/vggt">VGGT on GitHub</a></li>
             <li>🔗 <a href="https://github.com/xiongzhu666/Sky-Segmentation-and-Post-processing">ONNX for sky segmentation</a></li>
         </ul>
         <p>We extend our gratitude to these projects for their valuable contributions to the research community.</p>
@@ -200,38 +200,6 @@ def get_gradio_theme():
     )
 
 
-# Example scene thumbnail grid CSS (if needed separately)
-THUMBNAIL_CSS = """
-/* Make thumbnails clickable with pointer cursor */
-.clickable-thumbnail img {
-    cursor: pointer !important;
-}
-
-.clickable-thumbnail:hover img {
-    cursor: pointer !important;
-    opacity: 0.8;
-    transition: opacity 0.3s ease;
-}
-
-/* Make thumbnail containers narrower horizontally */
-.clickable-thumbnail {
-    padding: 5px 2px !important;
-    margin: 0 2px !important;
-}
-
-.clickable-thumbnail .image-container {
-    margin: 0 !important;
-    padding: 0 !important;
-}
-
-.scene-info {
-    text-align: center !important;
-    padding: 5px 2px !important;
-    margin: 0 !important;
-}
-"""
-
-
 # Measure tab instructions HTML
 MEASURE_INSTRUCTIONS_HTML = """
 ### Click on the image to measure the distance between two points.

mapanything/utils/hf_utils/hf_helpers.py

@@ -0,0 +1,186 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+"""
+Helper functions for HuggingFace integration and model initialization.
+"""
+
+import json
+import os
+
+
+def load_hf_token():
+    """Load HuggingFace access token from local file"""
+    token_file_paths = [
+        "/home/aknapitsch/hf_token.txt",
+    ]
+
+    for token_path in token_file_paths:
+        if os.path.exists(token_path):
+            try:
+                with open(token_path, "r") as f:
+                    token = f.read().strip()
+                print(f"Loaded HuggingFace token from: {token_path}")
+                return token
+            except Exception as e:
+                print(f"Error reading token from {token_path}: {e}")
+                continue
+        else:
+            print(token_path, "token_path doesnt exist")
+
+    # Also try environment variable
+    # see https://huggingface.co/docs/hub/spaces-overview#managing-secrets on options
+    token = (
+        os.getenv("HF_TOKEN")
+        or os.getenv("HUGGING_FACE_HUB_TOKEN")
+        or os.getenv("HUGGING_FACE_MODEL_TOKEN")
+    )
+    if token:
+        print("Loaded HuggingFace token from environment variable")
+        return token
+
+    print(
+        "Warning: No HuggingFace token found. Model loading may fail for private repositories."
+    )
+    return None
+
+
+def init_hydra_config(config_path, overrides=None):
+    """Initialize Hydra config"""
+    import hydra
+
+    config_dir = os.path.dirname(config_path)
+    config_name = os.path.basename(config_path).split(".")[0]
+    relative_path = os.path.relpath(config_dir, os.path.dirname(__file__))
+    hydra.core.global_hydra.GlobalHydra.instance().clear()
+    hydra.initialize(version_base=None, config_path=relative_path)
+    if overrides is not None:
+        cfg = hydra.compose(config_name=config_name, overrides=overrides)
+    else:
+        cfg = hydra.compose(config_name=config_name)
+    return cfg
+
+
+def initialize_mapanything_model(high_level_config, device):
+    """
+    Initialize MapAnything model with three-tier fallback approach:
+    1. Try HuggingFace from_pretrained()
+    2. Download HF config + use local model factory + load HF weights
+    3. Pure local configuration fallback
+
+    Args:
+        high_level_config (dict): Configuration dictionary containing model settings
+        device (torch.device): Device to load the model on
+
+    Returns:
+        torch.nn.Module: Initialized MapAnything model
+    """
+    import torch
+    from huggingface_hub import hf_hub_download
+    from mapanything.models import init_model, MapAnything
+
+    print("Initializing MapAnything model...")
+
+    # Initialize Hydra config and create model from configuration
+    cfg = init_hydra_config(
+        high_level_config["path"], overrides=high_level_config["config_overrides"]
+    )
+
+    # Try using from_pretrained first
+    try:
+        print("Loading MapAnything model from_pretrained...")
+        model = MapAnything.from_pretrained(high_level_config["hf_model_name"]).to(
+            device
+        )
+        print("Loading MapAnything model from_pretrained succeeded...")
+        return model
+    except Exception as e:
+        print(f"from_pretrained failed: {e}")
+        print("Falling back to local configuration approach using hf_hub_download...")
+
+        # Create model from local configuration instead of using from_pretrained
+        # Try to download and use the config from HuggingFace Hub
+        try:
+            print("Downloading model configuration from HuggingFace Hub...")
+            config_path = hf_hub_download(
+                repo_id=high_level_config["hf_model_name"],
+                filename=high_level_config["config_name"],
+                token=load_hf_token(),
+            )
+
+            # Load the config from the downloaded file
+            with open(config_path, "r") as f:
+                downloaded_config = json.load(f)
+
+            print("Using downloaded configuration for model initialization")
+            model = init_model(
+                model_str=downloaded_config.get(
+                    "model_str", high_level_config["model_str"]
+                ),
+                model_config=downloaded_config.get(
+                    "model_config", cfg.model.model_config
+                ),
+                torch_hub_force_reload=high_level_config.get(
+                    "torch_hub_force_reload", False
+                ),
+            )
+        except Exception as config_e:
+            print(f"Failed to download/use HuggingFace config: {config_e}")
+            print("Falling back to local configuration...")
+            # Fall back to local configuration as before
+            model = init_model(
+                model_str=cfg.model.model_str,
+                model_config=cfg.model.model_config,
+                torch_hub_force_reload=high_level_config.get(
+                    "torch_hub_force_reload", False
+                ),
+            )
+
+        # Load the pretrained weights from HuggingFace Hub
+        try:
+            # First, let's see what files are available in the repository
+            try:
+                checkpoint_filename = high_level_config["checkpoint_name"]
+                # Download the model weights
+                checkpoint_path = hf_hub_download(
+                    repo_id=high_level_config["hf_model_name"],
+                    filename=checkpoint_filename,
+                    token=load_hf_token(),
+                )
+
+                # Load the weights
+                print("start loading checkpoint")
+                if checkpoint_filename.endswith(".safetensors"):
+                    from safetensors.torch import load_file
+
+                    checkpoint = load_file(checkpoint_path)
+                else:
+                    checkpoint = torch.load(
+                        checkpoint_path, map_location="cpu", weights_only=False
+                    )
+
+                print("start loading state_dict")
+                if "model" in checkpoint:
+                    model.load_state_dict(checkpoint["model"], strict=False)
+                elif "state_dict" in checkpoint:
+                    model.load_state_dict(checkpoint["state_dict"], strict=False)
+                else:
+                    model.load_state_dict(checkpoint, strict=False)
+
+                print(
+                    f"Successfully loaded pretrained weights from HuggingFace Hub ({checkpoint_filename})"
+                )
+
+            except Exception as inner_e:
+                print(f"Error listing repository files or loading weights: {inner_e}")
+                raise inner_e
+
+        except Exception as e:
+            print(f"Warning: Could not load pretrained weights: {e}")
+            print("Proceeding with randomly initialized model...")
+
+        model = model.to(device)
+        return model

requirements.txt

@@ -1,130 +1,22 @@
-aiofiles==24.1.0
-annotated-types==0.7.0
-antlr4-python3-runtime==4.9.3
-anyio==4.10.0
-argon2-cffi==25.1.0
-argon2-cffi-bindings==25.1.0
-asttokens==3.0.0
-attrs==25.3.0
-Brotli==1.1.0
-certifi==2025.8.3
-cffi==1.17.1
-charset-normalizer==3.4.3
-click==8.2.1
-coloredlogs==15.0.1
-contourpy==1.3.2
-cycler==0.12.1
-decorator==5.2.1
-einops==0.8.1
-exceptiongroup==1.3.0
-executing==2.2.1
-fastapi==0.116.1
-ffmpy==0.6.1
-filelock==3.19.1
-flatbuffers==25.2.10
-fonttools==4.59.2
-fsspec==2025.9.0
-gradio==5.44.1
-gradio_client==1.12.1
-groovy==0.1.2
-h11==0.16.0
-hf-xet==1.1.9
-httpcore==1.0.9
-httpx==0.28.1
-huggingface-hub==0.34.4
-humanfriendly==10.0
-hydra-core==1.3.2
-idna==3.10
-ipdb==0.13.13
-ipython==8.37.0
-jaxtyping==0.3.2
-jedi==0.19.2
-Jinja2==3.1.6
-joblib==1.5.2
-kiwisolver==1.4.9
-markdown-it-py==4.0.0
-MarkupSafe==3.0.2
-matplotlib==3.10.6
-matplotlib-inline==0.1.7
-mdurl==0.1.2
-minio==7.2.16
-mpmath==1.3.0
-networkx==3.4.2
-numpy==2.2.6
-nvidia-cublas-cu12==12.8.4.1
-nvidia-cuda-cupti-cu12==12.8.90
-nvidia-cuda-nvrtc-cu12==12.8.93
-nvidia-cuda-runtime-cu12==12.8.90
-nvidia-cudnn-cu12==9.10.2.21
-nvidia-cufft-cu12==11.3.3.83
-nvidia-cufile-cu12==1.13.1.3
-nvidia-curand-cu12==10.3.9.90
-nvidia-cusolver-cu12==11.7.3.90
-nvidia-cusparse-cu12==12.5.8.93
-nvidia-cusparselt-cu12==0.7.1
-nvidia-nccl-cu12==2.27.3
-nvidia-nvjitlink-cu12==12.8.93
-nvidia-nvtx-cu12==12.8.90
-omegaconf==2.3.0
-onnxruntime==1.22.1
-opencv-python-headless==4.12.0.88
-orjson==3.11.3
-packaging==25.0
-pandas==2.3.2
-parso==0.8.5
-pexpect==4.9.0
-pillow==11.3.0
-prompt_toolkit==3.0.52
-protobuf==6.32.0
-psutil==5.9.8
-ptyprocess==0.7.0
-pure_eval==0.2.3
-pyarrow==21.0.0
-pycparser==2.22
-pycryptodome==3.23.0
-pydantic==2.11.7
-pydantic_core==2.33.2
-pydub==0.25.1
-Pygments==2.19.2
-pyparsing==3.2.3
-python-dateutil==2.9.0.post0
-python-multipart==0.0.20
-pytz==2025.2
-PyYAML==6.0.2
-requests==2.32.5
-rerun-sdk==0.24.1
-rich==14.1.0
-ruff==0.12.11
-safehttpx==0.1.6
-safetensors==0.6.2
-scikit-learn==1.7.1
-scipy==1.15.3
-semantic-version==2.10.0
-shellingham==1.5.4
-six==1.17.0
-sniffio==1.3.1
-spaces==0.40.1
-stack-data==0.6.3
-starlette==0.47.3
-sympy==1.14.0
-threadpoolctl==3.6.0
-tomli==2.2.1
-tomlkit==0.13.3
-torch==2.8.0
-torchaudio==2.8.0
-torchvision==0.23.0
-tqdm==4.67.1
-traitlets==5.14.3
-trimesh==4.8.1
-triton==3.4.0
-turm==0.7.3
-typer==0.17.3
-typing-inspection==0.4.1
-typing_extensions==4.15.0
-tzdata==2025.2
-uniception==0.1.4
-urllib3==2.5.0
-uvicorn==0.35.0
-wadler_lindig==0.1.7
-wcwidth==0.2.13
-websockets==15.0.1
+--extra-index-url https://download.pytorch.org/whl/cu113
+torch
+torchvision
+torchaudio
+gradio
+huggingface-hub
+numpy
+opencv-python-headless
+Pillow
+matplotlib
+scikit-learn
+scipy
+spaces
+hydra-core
+omegaconf
+trimesh
+einops
+requests
+psutil
+tqdm
+safetensors
+uniception==0.1.4