make the dockerfile reference the fork of flat-bug

Author: Atticus29

processing_services/example/FLAT_BUG_IMPLEMENTATION.md

@@ -7,31 +7,35 @@ I've successfully implemented the `FlatBugObjectDetector` class to use the actua
 ## Key Changes Made
 
 ### 1. Updated `compile()` method
+
 - **Before**: Used `transformers.pipeline` with a placeholder checkpoint
-- **After**: Uses `flat_bug.predictor.Predictor` with the default model `'flat_bug_M.pt'`
+- **After**: Uses `flat_bug.predictor.Predictor` with the default model `'flat_bug_N.pt'`
 - The model will be automatically downloaded on first use
 - Added configurable hyperparameters (score threshold, IoU threshold, etc.)
 
 ### 2. Updated `run()` method
+
 - **Before**: Called `self.model(image, candidate_labels=...)`
 - **After**: Uses `self.model.pyramid_predictions(image)` which is the flat-bug API
 - Handles the `TensorPredictions` response format from flat-bug
 - Converts tensors to numpy arrays and extracts bounding boxes and scores
 
 ### 3. Updated description
+
 - Now accurately reflects that it uses the actual flat-bug library
 - Mentions specialization for terrestrial arthropod detection
 
 ## How It Works
 
 1. **Installation**: Flat-bug needs to be installed from source:
+
    ```bash
    git clone https://github.com/darsa-group/flat-bug.git
    cd flat-bug
    pip install -e .
    ```
 
-2. **Model Loading**: The `flat_bug_M.pt` model is downloaded automatically on first use
+2. **Model Loading**: The `flat_bug_N.pt` model is downloaded automatically on first use
 
 3. **Inference**: Uses flat-bug's pyramid tiling approach for detection on arbitrarily large images
 
@@ -50,6 +54,7 @@ pip install torch>=2.3
 ## Testing
 
 I've created `test_flat_bug_implementation.py` which you can run to:
+
 - Verify the flat-bug installation
 - Inspect the actual format of `TensorPredictions` objects
 - Confirm the attribute names and data structures

processing_services/example/api/algorithms.py

@@ -199,7 +199,7 @@ def compile(self, device: str | None = None):
             logger.info(f"Compiling {self.algorithm_config_response.name} on device {device_choice}...")
 
             # Initialize flat-bug predictor with default model
-            self.model = Predictor(model="flat_bug_M.pt", device=device_choice)  # Default flat-bug model
+            self.model = Predictor(model="flat_bug_N.pt", device=device_choice)  # Default flat-bug model
 
             # Set some reasonable hyperparameters
             # TIME=False is critical to avoid CUDA event errors when running on CPU

processing_services/example/requirements.txt

@@ -9,5 +9,5 @@ torchvision==0.21.0
 scipy==1.16.0
 # Use headless OpenCV to avoid GUI dependencies in Docker
 opencv-python-headless==4.13.0.92
-flat_bug @ https://github.com/darsa-group/flat-bug/archive/refs/heads/main.zip
+flat_bug @ https://github.com/atticus29/flat-bug/archive/refs/heads/main.zip
 

processing_services/example/test_flat_bug_implementation.py

@@ -36,7 +36,7 @@ def test_flat_bug_api():
 
         # Initialize predictor
         device = "cuda" if torch.cuda.is_available() else "cpu"
-        predictor = Predictor(model="flat_bug_M.pt", device=device)
+        predictor = Predictor(model="flat_bug_N.pt", device=device)
         print(f"✓ Initialized Predictor with device: {device}")
 
         # Set hyperparameters - DISABLE timing to avoid CUDA event error on CPU
@@ -106,7 +106,7 @@ def show_installation_instructions():
     print("3. Make sure PyTorch is installed:")
     print("   pip install torch>=2.3")
     print()
-    print("4. The model 'flat_bug_M.pt' will be downloaded automatically")
+    print("4. The model 'flat_bug_N.pt' will be downloaded automatically")
     print("   on first use.")
     print()