Create app.py

app.py

@@ -0,0 +1,176 @@
+import os
+import shutil
+import zipfile
+import pathlib
+import tempfile
+import gradio
+import pandas
+import PIL.Image
+import huggingface_hub
+import autogluon.multimodal
+
+# Model configuration
+MODEL_REPO_ID = "Anyuhhh/sign-language-recognition"  # Your Hugging Face Space repo
+ZIP_FILENAME = "autogluon_image_predictor_dir.zip"
+HF_TOKEN = os.getenv("HF_TOKEN", None)  # Optional: only if your repo is private
+
+CACHE_DIR = pathlib.Path("hf_assets")
+EXTRACT_DIR = CACHE_DIR / "predictor_native"
+
+def _prepare_predictor_dir() -> str:
+    """Download and extract the predictor from Hugging Face"""
+    CACHE_DIR.mkdir(parents=True, exist_ok=True)
+    
+    print(f"Downloading model from HuggingFace: {MODEL_REPO_ID}/{ZIP_FILENAME}")
+    
+    try:
+        # Download from Hugging Face
+        local_zip = huggingface_hub.hf_hub_download(
+            repo_id=MODEL_REPO_ID,
+            filename=ZIP_FILENAME,
+            repo_type="space",  # Changed to "space" since you're uploading to your Space repo
+            token=HF_TOKEN,
+            local_dir=str(CACHE_DIR),
+            local_dir_use_symlinks=False,
+        )
+        print(f"Downloaded to: {local_zip}")
+    except Exception as e:
+        print(f"Error downloading from Space repo: {e}")
+        print("Trying as model repo instead...")
+        # Fallback: try as model repo
+        local_zip = huggingface_hub.hf_hub_download(
+            repo_id=MODEL_REPO_ID,
+            filename=ZIP_FILENAME,
+            repo_type="model",
+            token=HF_TOKEN,
+            local_dir=str(CACHE_DIR),
+            local_dir_use_symlinks=False,
+        )
+    
+    # Clean and recreate extraction directory
+    if EXTRACT_DIR.exists():
+        shutil.rmtree(EXTRACT_DIR)
+    EXTRACT_DIR.mkdir(parents=True, exist_ok=True)
+    
+    # Extract the ZIP file
+    print(f"Extracting to: {EXTRACT_DIR}")
+    with zipfile.ZipFile(local_zip, "r") as zf:
+        zf.extractall(str(EXTRACT_DIR))
+    
+    # Find the predictor directory
+    contents = list(EXTRACT_DIR.iterdir())
+    predictor_root = contents[0] if (len(contents) == 1 and contents[0].is_dir()) else EXTRACT_DIR
+    
+    print(f"Predictor directory: {predictor_root}")
+    return str(predictor_root)
+
+# Initialize predictor
+print("Initializing predictor...")
+PREDICTOR_DIR = _prepare_predictor_dir()
+PREDICTOR = autogluon.multimodal.MultiModalPredictor.load(PREDICTOR_DIR)
+print("Predictor loaded successfully!")
+
+# Sign language classes A-Z (26 classes)
+CLASS_LABELS = {i: chr(65+i) for i in range(26)}  # 0='A', 1='B', ..., 25='Z'
+
+def _human_label(c):
+    """Convert class index to human-readable label"""
+    try:
+        ci = int(c)
+        return CLASS_LABELS.get(ci, str(c))
+    except Exception:
+        return CLASS_LABELS.get(c, str(c))
+
+def do_predict(pil_img: PIL.Image.Image):
+    """Predict sign language letter from image"""
+    if pil_img is None:
+        return {}, None
+
+    tmpdir = pathlib.Path(tempfile.mkdtemp())
+    img_path = tmpdir / "input.png"
+
+    # Preprocess image - resize and convert to RGB
+    if pil_img.mode != 'RGB':
+        pil_img = pil_img.convert('RGB')
+    processed_img = pil_img.resize((224, 224))
+    processed_img.save(img_path)
+
+    # Create DataFrame for prediction
+    df = pandas.DataFrame({"image": [str(img_path)]})
+
+    # Get predictions
+    proba_df = PREDICTOR.predict_proba(df)
+
+    # Create pretty labels with probabilities
+    pretty_dict = {}
+    for col in proba_df.columns:
+        if isinstance(col, int) and col < 26:
+            label = f"Letter {CLASS_LABELS[col]}"
+        else:
+            label = str(col)
+        pretty_dict[label] = float(proba_df[col].iloc[0])
+
+    # Sort by probability (highest first)
+    pretty_dict = dict(sorted(pretty_dict.items(), key=lambda x: x[1], reverse=True))
+
+    # Cleanup
+    shutil.rmtree(tmpdir, ignore_errors=True)
+
+    return pretty_dict, processed_img
+
+# Example sign language images
+EXAMPLES = [
+    ["https://www.signingsavvy.com/images/words/alphabet/2/a1.jpg"],
+    ["https://www.signingsavvy.com/images/words/alphabet/2/b1.jpg"],
+    ["https://www.signingsavvy.com/images/words/alphabet/2/c1.jpg"]
+]
+
+# Gradio UI
+with gradio.Blocks(theme=gradio.themes.Soft()) as demo:
+    gradio.Markdown("# 🤟 Sign Language Recognition")
+    gradio.Markdown("""
+    This app uses an AutoGluon multimodal predictor to recognize American Sign Language (ASL) letters.
+    
+    **How to use:**
+    1. Upload a photo of a hand sign or use your webcam
+    2. The model will predict which letter (A-Z) it represents
+    3. View the top 5 predictions with confidence scores
+    """)
+
+    with gradio.Row():
+        with gradio.Column():
+            image_in = gradio.Image(
+                type="pil", 
+                label="Upload hand sign image", 
+                sources=["upload", "webcam"]
+            )
+        with gradio.Column():
+            processed_out = gradio.Image(
+                type="pil", 
+                label="Preprocessed image (what model sees - 224x224)"
+            )
+
+    proba_pretty = gradio.Label(num_top_classes=5, label="Top 5 predictions")
+
+    # Update on image change
+    image_in.change(
+        fn=do_predict, 
+        inputs=[image_in], 
+        outputs=[proba_pretty, processed_out]
+    )
+
+    gradio.Examples(
+        examples=EXAMPLES,
+        inputs=[image_in],
+        label="Example ASL signs (click to try)",
+        examples_per_page=3,
+        cache_examples=False,
+    )
+
+    gradio.Markdown("""
+    ---
+    **Note:** This model recognizes static ASL letters (A-Z). For best results, use clear images with good lighting.
+    """)
+
+if __name__ == "__main__":
+    demo.launch()