fix: sam2 (HumanSignal#663)

Co-authored-by: LI, Gaoyang <[email protected]>

Author: micaelakaplan

label_studio_ml/examples/segment_anything_2_image/README.md

@@ -133,16 +133,27 @@ cd label_studio_ml/examples/segment_anything_2_image
 pip install -r requirements.txt
 ```
 
-2. Download [`segment-anything-2` repo](https://github.com/facebookresearch/segment-anything-2) into the root directory. Install SegmentAnything model and download checkpoints using [the official Meta documentation](https://github.com/facebookresearch/segment-anything-2?tab=readme-ov-file#installation)
-
+2. Download [`segment-anything-2` repo](https://github.com/facebookresearch/sam2) into the root directory. Install SegmentAnything model and download checkpoints using [the official Meta documentation](https://github.com/facebookresearch/sam2?tab=readme-ov-file#installation)
+You should now have the following folder structure: 
+
+    | root directory 
+        | label-studio-ml-backend 
+            | label-studio-ml
+                | examples 
+                    | segment_anythng_2_image
+        | sam2
+            | sam2
+            | checkpoints
 
 3. Then you can start the ML backend on the default port `9090`:
 
 ```bash
-cd ../
-label-studio-ml start ./segment_anything_2_image
+cd ~/sam2
+label-studio-ml start ../label-studio-ml-backend/label_studio_ml/examples/segment_anything_2_image
 ```
 
+Due to breaking changes from Meta [HERE](https://github.com/facebookresearch/sam2/blob/c2ec8e14a185632b0a5d8b161928ceb50197eddc/sam2/build_sam.py#L20), it is CRUCIAL that you run this command from the sam2 directory at your root directory. 
+
 4. Connect running ML backend server to Label Studio: go to your project `Settings -> Machine Learning -> Add Model` and specify `http://localhost:9090` as a URL. Read more in the official [Label Studio documentation](https://labelstud.io/guide/ml#Connect-the-model-to-Label-Studio).
 
 ## Running with Docker (coming soon)

label_studio_ml/examples/segment_anything_2_image/model.py

@@ -1,6 +1,7 @@
 import torch
 import numpy as np
 import os
+import sys
 import pathlib
 from typing import List, Dict, Optional
 from uuid import uuid4
@@ -9,14 +10,16 @@
 from label_studio_sdk.converter import brush
 from label_studio_sdk._extensions.label_studio_tools.core.utils.io import get_local_path
 from PIL import Image
+
+ROOT_DIR = os.getcwd()
+sys.path.insert(0, ROOT_DIR)
 from sam2.build_sam import build_sam2
 from sam2.sam2_image_predictor import SAM2ImagePredictor
 
 
 DEVICE = os.getenv('DEVICE', 'cuda')
-SEGMENT_ANYTHING_2_REPO_PATH = os.getenv('SEGMENT_ANYTHING_2_REPO_PATH', 'segment-anything-2')
-MODEL_CONFIG = os.getenv('MODEL_CONFIG', 'sam2_hiera_l.yaml')
-MODEL_CHECKPOINT = os.getenv('MODEL_CHECKPOINT', 'sam2_hiera_large.pt')
+MODEL_CONFIG = os.getenv('MODEL_CONFIG', 'configs/sam2.1/sam2.1_hiera_l.yaml')
+MODEL_CHECKPOINT = os.getenv('MODEL_CHECKPOINT', 'sam2.1_hiera_large.pt')
 
 if DEVICE == 'cuda':
     # use bfloat16 for the entire notebook
@@ -29,7 +32,7 @@
 
 
 # build path to the model checkpoint
-sam2_checkpoint = str(pathlib.Path(__file__).parent / SEGMENT_ANYTHING_2_REPO_PATH / "checkpoints" / MODEL_CHECKPOINT)
+sam2_checkpoint = str(os.path.join(ROOT_DIR, "checkpoints", MODEL_CHECKPOINT))
 
 sam2_model = build_sam2(MODEL_CONFIG, sam2_checkpoint, device=DEVICE)