Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+model18cls/saved_model/variables/variables.data-00000-of-00001 filter=lfs diff=lfs merge=lfs -text

README.md

@@ -13,6 +13,23 @@ license: mit
 
 Fast GPU-accelerated brain MRI segmentation API.
 
+## Setup Instructions
+
+### 1. Convert the model (run locally first)
+
+```bash
+# Install conversion dependencies
+pip install tensorflowjs tensorflow
+
+# Convert tfjs model to H5 format
+cd huggingface
+python convert_model.py ../public/models/model18cls ./model18cls
+```
+
+### 2. Upload to Hugging Face Space
+
+The `model18cls/` folder should now contain `model.h5` and `saved_model/`.
+
 ## API Endpoints
 
 ### POST /segment

app.py

@@ -25,31 +25,41 @@ app.add_middleware(
 model = None
 MODEL_PATH = "model18cls"
 
-def load_tfjs_model(model_path):
-    """
-    Load a TensorFlow.js LayersModel from model.json + weight shards.
-    Converts it to a Keras model for inference.
+def load_model():
     """
-    import tensorflowjs as tfjs
-
-    # Load the tfjs model and convert to Keras
-    model = tfjs.converters.load_keras_model(os.path.join(model_path, "model.json"))
-    return model
+    Load TensorFlow model on startup.
+    Supports H5, SavedModel, or Keras formats.
 
-def load_model():
-    """Load TensorFlow model on startup"""
+    NOTE: Convert tfjs models first using convert_model.py
+    """
     global model
     if model is None:
         print(f"Loading model from {MODEL_PATH}...")
 
-        # Check if it's a tfjs model (has model.json) or SavedModel
-        model_json_path = os.path.join(MODEL_PATH, "model.json")
-        if os.path.exists(model_json_path):
-            print("Detected TensorFlow.js format, converting...")
-            model = load_tfjs_model(MODEL_PATH)
-        else:
-            print("Loading as SavedModel format...")
+        # Check for different model formats
+        h5_path = os.path.join(MODEL_PATH, "model.h5")
+        keras_path = os.path.join(MODEL_PATH, "model.keras")
+        saved_model_dir = os.path.join(MODEL_PATH, "saved_model")
+
+        if os.path.exists(h5_path):
+            print("Loading H5 format...")
+            model = tf.keras.models.load_model(h5_path)
+        elif os.path.exists(keras_path):
+            print("Loading Keras format...")
+            model = tf.keras.models.load_model(keras_path)
+        elif os.path.exists(saved_model_dir):
+            print("Loading SavedModel format...")
+            model = tf.keras.models.load_model(saved_model_dir)
+        elif os.path.exists(MODEL_PATH) and os.path.isdir(MODEL_PATH):
+            # Try loading directory as SavedModel
+            print("Loading as SavedModel directory...")
             model = tf.keras.models.load_model(MODEL_PATH)
+        else:
+            raise FileNotFoundError(
+                f"No model found in {MODEL_PATH}. "
+                "Please convert the tfjs model first using: "
+                "python convert_model.py ../public/models/model18cls ./model18cls"
+            )
 
         print("Model loaded successfully!")
         print(f"Input shape: {model.input_shape}")

convert_model.py

@@ -0,0 +1,52 @@
+#!/usr/bin/env python3
+"""
+Convert TensorFlow.js model to Keras H5 format.
+Run this locally before uploading to Hugging Face.
+
+Usage:
+    pip install tensorflowjs tensorflow
+    python convert_model.py ../public/models/model18cls ./model18cls
+"""
+
+import sys
+import os
+
+def convert_tfjs_to_keras(input_path, output_path):
+    """Convert tfjs model to Keras H5 format"""
+    # Import here to avoid issues if not installed
+    import tensorflowjs as tfjs
+    import tensorflow as tf
+
+    print(f"Converting {input_path} to Keras format...")
+
+    # Load tfjs model
+    model_json = os.path.join(input_path, "model.json")
+    model = tfjs.converters.load_keras_model(model_json)
+
+    # Create output directory
+    os.makedirs(output_path, exist_ok=True)
+
+    # Save as H5
+    h5_path = os.path.join(output_path, "model.h5")
+    model.save(h5_path)
+    print(f"Saved to {h5_path}")
+
+    # Also save as SavedModel for better compatibility
+    savedmodel_path = os.path.join(output_path, "saved_model")
+    model.save(savedmodel_path, save_format='tf')
+    print(f"Saved to {savedmodel_path}")
+
+    print("Conversion complete!")
+    print(f"Model input shape: {model.input_shape}")
+    print(f"Model output shape: {model.output_shape}")
+
+if __name__ == "__main__":
+    if len(sys.argv) < 3:
+        print("Usage: python convert_model.py <input_tfjs_path> <output_path>")
+        print("Example: python convert_model.py ../public/models/model18cls ./model18cls")
+        sys.exit(1)
+
+    input_path = sys.argv[1]
+    output_path = sys.argv[2]
+
+    convert_tfjs_to_keras(input_path, output_path)

model18cls/model.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:79ec7062d63fc42594a0c6538247934b97c7042d70033760ca688cbf99cb0d34
+size 442236

model18cls/saved_model/fingerprint.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c960c45c2b00bf76cfa8d6ac4924a33a5be27b2fb4278543b048ba7740ce3bd
+size 77

model18cls/saved_model/keras_metadata.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb6956e1d4160290701b3e5bda555c1a679b6be456f749ac22999276562d1ee5
+size 40634

model18cls/saved_model/saved_model.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:11505c3a150ae680e364454caf30210518f633ae92b80b0b178725c16fa2afde
+size 242134

model18cls/saved_model/variables/variables.data-00000-of-00001

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a11f9621a21dc7a3f50f28c8d575d943b94c481ed9604e5815031952147ce08
+size 399788

requirements.txt

@@ -2,6 +2,5 @@ fastapi==0.104.1
 uvicorn[standard]==0.24.0
 python-multipart==0.0.6
 tensorflow==2.15.0
-tensorflowjs==4.15.0
 nibabel==5.2.0
 numpy==1.26.2