keep only edgetam

README.md

@@ -1,5 +1,5 @@
 ---
-title: Segment Anything 2 Video Tracking
+title: EdgeTAM Video Tracking
 emoji: 👀
 colorFrom: purple
 colorTo: indigo
@@ -8,7 +8,7 @@ sdk_version: 5.47.2
 app_file: app.py
 pinned: false
 license: apache-2.0
-short_description: Segment any objects and track them through a video with SAM2
+short_description: Segment and track any objects with EdgeTAM
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -122,7 +122,7 @@ class AppState:
         self.pending_box_start_obj_id: int | None = None
         self.is_switching_model: bool = False
         # Model selection
-        self.model_repo_key: str = "tiny"
+        self.model_repo_key: str = "EdgeTAM"
         self.model_repo_id: str | None = None
         self.session_repo_id: str | None = None
 
@@ -135,14 +135,8 @@ class AppState:
 
 
 def _model_repo_from_key(key: str) -> str:
-    mapping = {
-        "tiny": "facebook/sam2.1-hiera-tiny",
-        "small": "facebook/sam2.1-hiera-small",
-        "base_plus": "facebook/sam2.1-hiera-base-plus",
-        "large": "facebook/sam2.1-hiera-large",
-        "EdgeTAM": "yonigozlan/EdgeTAM-hf",
-    }
-    return mapping.get(key, mapping["base_plus"])
+    mapping = {"EdgeTAM": "yonigozlan/EdgeTAM-hf"}
+    return mapping.get(key)
 
 
 def load_model_if_needed(GLOBAL_STATE: gr.State) -> tuple[AutoModel, Sam2VideoProcessor, str, torch.dtype]:
@@ -533,13 +527,13 @@ def reset_session(GLOBAL_STATE: gr.State) -> tuple[AppState, Image.Image, int, i
 
 theme = Soft(primary_hue="blue", secondary_hue="rose", neutral_hue="slate")
 
-with gr.Blocks(title="SAM2 Video (Transformers) - Interactive Segmentation", theme=theme) as demo:
+with gr.Blocks(title="EdgeTAM Video (Transformers) - Interactive Segmentation", theme=theme) as demo:
     GLOBAL_STATE = gr.State(AppState())
 
     gr.Markdown(
         """
-        ### SAM2 Video Tracking · powered by Hugging Face 🤗 Transformers
-        Segment and track objects across a video with SAM2 (Segment Anything 2). This demo runs the official implementation from the Hugging Face Transformers library for interactive, promptable video segmentation.
+        ### EdgeTAM Video Tracking · powered by Hugging Face 🤗 Transformers
+        Segment and track objects across a video with EdgeTAM. This demo runs the official implementation from the Hugging Face Transformers library for interactive, promptable video segmentation.
         """
     )
     with gr.Row():
@@ -548,7 +542,6 @@ with gr.Blocks(title="SAM2 Video (Transformers) - Interactive Segmentation", the
                 """
                 **Quick start**
                 - **Load a video**: Upload your own or pick an example below.
-                - **Checkpoint**: Tiny / Small / Base+ / Large (trade speed vs. accuracy).
                 - **Points mode**: Select an Object ID and point label (positive/negative), then click the frame to add guidance. You can add **multiple points per object** and define **multiple objects** across frames.
                 - **Boxes mode**: Click two opposite corners to draw a box. Old inputs for that object are cleared automatically.
                 """
@@ -560,19 +553,13 @@ with gr.Blocks(title="SAM2 Video (Transformers) - Interactive Segmentation", the
                 - **Preview**: Use the slider to navigate frames and see the current masks.
                 - **Propagate**: Click “Propagate across video” to track all defined objects through the entire video. The preview follows progress periodically to keep things responsive.
                 - **Export**: Render an MP4 for smooth playback using the original video FPS.
-                - **Note**: More info on the Hugging Face 🤗 Transformers implementation of SAM2 can be found [here](https://huggingface.co/docs/transformers/en/main/en/model_doc/sam2_video).
+                - **Note**: More info on the Hugging Face 🤗 Transformers implementation of EdgeTAM can be found [here](https://huggingface.co/docs/transformers/en/main/en/model_doc/edgetam_video).
                 """
             )
 
     with gr.Row():
         with gr.Column(scale=1):
             video_in = gr.Video(label="Upload video", sources=["upload", "webcam"], interactive=True)
-            ckpt_radio = gr.Radio(
-                choices=["tiny", "small", "base_plus", "large", "EdgeTAM"],
-                value="tiny",
-                label="SAM2.1 checkpoint",
-            )
-            ckpt_progress = gr.Markdown(visible=False)
             load_status = gr.Markdown(visible=True)
             reset_btn = gr.Button("Reset Session", variant="secondary")
         with gr.Column(scale=2):
@@ -628,26 +615,6 @@ with gr.Blocks(title="SAM2 Video (Transformers) - Interactive Segmentation", the
         render_btn = gr.Button("Render MP4 for smooth playback", variant="primary")
     playback_video = gr.Video(label="Rendered Playback", interactive=False)
 
-    def _on_ckpt_change(s: AppState, key: str):
-        if s is not None and key:
-            key = str(key)
-            if key != s.model_repo_key:
-                # Update and drop current model to reload lazily next time
-                s.is_switching_model = True
-                s.model_repo_key = key
-                s.model_repo_id = None
-                s.model = None
-                s.processor = None
-        # Stream progress text while loading (first yield shows text)
-        yield gr.update(visible=True, value=f"Loading checkpoint: {key}...")
-        ensure_session_for_current_model(s)
-        if s is not None:
-            s.is_switching_model = False
-        # Final yield hides the text
-        yield gr.update(visible=False, value="")
-
-    ckpt_radio.change(_on_ckpt_change, inputs=[GLOBAL_STATE, ckpt_radio], outputs=[ckpt_progress])
-
     def _sync_frame_idx(state_in: AppState, idx: int):
         if state_in is not None:
             state_in.current_frame_idx = int(idx)