app.py

import gradio_client.utils as _cu

# Fix for gradio-client bug where additionalProperties=True causes a crash
_orig_parse = _cu._json_schema_to_python_type
def _safe_parse(schema, defs=None):
    if not isinstance(schema, dict):
        return "any"
    return _orig_parse(schema, defs)
_cu._json_schema_to_python_type = _safe_parse

import gradio as gr
import numpy as np
import os
from PIL import Image

EMOTIONS = ["Angry", "Disgust", "Fear", "Happy", "Neutral", "Sad", "Surprised"]

model = None

def load_model():
    global model
    try:
        # Use tf_keras (legacy Keras 2) to load the old Keras 2.5 model format
        import tf_keras
        from tf_keras.models import model_from_json

        if not os.path.exists("model.json") or not os.path.exists("best.h5"):
            print("⚠️ model.json or best.h5 not found.")
            return

        with open("model.json", "r") as f:
            model_json = f.read()

        model = model_from_json(model_json)
        model.load_weights("best.h5")
        print("✅ Model loaded successfully.")
    except Exception as e:
        print(f"Model not loaded: {e}")

load_model()

def predict(image):
    if model is None:
        return "⚠️ Model could not be loaded.", {}
    if image is None:
        return "⚠️ Please upload an image first.", {}

    image = image.convert("L").resize((48, 48))
    arr = np.array(image, dtype=np.float32) / 255.0
    arr = arr.reshape(1, 48, 48, 1)

    preds = model.predict(arr, verbose=0)[0]
    scores = {e: round(float(p), 4) for e, p in zip(EMOTIONS, preds)}
    top = max(scores, key=scores.get)
    return f"{top} ({scores[top]:.0%} confidence)", scores

demo = gr.Interface(
    fn=predict,
    inputs=gr.Image(type="pil", label="📷 Upload face image"),
    outputs=[
        gr.Textbox(label="Prediction"),
        gr.JSON(label="Confidence scores"),
    ],
    title="🙂 Facial Expression Recognition",
    description="**CNN · 7 emotion classes · TensorFlow/Keras**\n\nUpload a face image to predict the emotion.",
    flagging_mode="never",
)

demo.launch()