import gradio as gr from ultralytics import YOLO from PIL import Image import numpy as np # ===================================== # LOAD MODEL # ===================================== MODEL_PATH = "best.pt" model = YOLO(MODEL_PATH) CLASS_NAMES = model.names # {0: 'paregi', 1: 'pelise'} IMG_SIZE = 640 CONF_THRESH = 0.15 print("✅ YOLOv8 model loaded") # ===================================== # PREDICTION FUNCTION # ===================================== def predict(image: Image.Image): if image is None: return {"error": "No image provided"} # Run inference results = model.predict( source=image, imgsz=IMG_SIZE, conf=CONF_THRESH, verbose=False ) r = results[0] if len(r.boxes) == 0: return { "prediction": "no_object_detected", "confidence": 0.0, "probabilities": { CLASS_NAMES[0]: 0.0, CLASS_NAMES[1]: 0.0 } } # Pick highest-confidence detection idx = r.boxes.conf.argmax() cls_id = int(r.boxes.cls[idx]) conf = float(r.boxes.conf[idx]) probs = { CLASS_NAMES[0]: 0.0, CLASS_NAMES[1]: 0.0 } probs[CLASS_NAMES[cls_id]] = conf return { "prediction": CLASS_NAMES[cls_id], "confidence": round(conf, 4), "probabilities": probs } # ===================================== # GRADIO UI # ===================================== app = gr.Interface( fn=predict, inputs=gr.Image(type="pil", label="Upload O-Ring Image"), outputs=gr.JSON(label="Prediction"), title="ann2ann – Paregi / Pelise Detection", description=( "YOLOv8 model trained to classify O-ring defects.\n\n" "Classes:\n" "- paregi\n" "- pelise\n\n" "Input images are resized to 640×640 automatically." ), examples=None ) # ===================================== # LAUNCH # ===================================== if __name__ == "__main__": app.launch()