from pathlib import Path import cv2 import gradio as gr import mediapipe as mp import torch from . import test from .hand_tracker import clean_coordinates_for_csv, clean_data hand_detector = mp.solutions.hands.Hands( static_image_mode=True, max_num_hands=1, min_detection_confidence=0.5, min_tracking_confidence=0.5 ) device = "cpu" checkpoint_path = Path(__file__).resolve().parent / "checkpoints" / "hand_gesture_model.pth" model = test.load_model(str(checkpoint_path)).to(device) def predict_num(image): if image is None: return None, {"No image provided": 1.0} processed_hand = hand_detector.process(image) if processed_hand.multi_hand_landmarks: for hand_landmarks in processed_hand.multi_hand_landmarks: landmarks = clean_data(1, hand_landmarks.landmark) coordinates = clean_coordinates_for_csv(landmarks) x = torch.tensor(coordinates, dtype=torch.float32).reshape(1, -1).to(device) logits = model(x) probs = torch.softmax(logits, dim=1) mp.solutions.drawing_utils.draw_landmarks( image, hand_landmarks, mp.solutions.hands.HAND_CONNECTIONS, ) prob_dict = {str(i): float(probs.reshape(-1)[i - 1].item()) for i in range(1, 6)} return image, prob_dict return image, {"No hand detected": 1.0} demo = gr.Interface( fn=predict_num, inputs=gr.Image(sources=["webcam"], type="numpy"), outputs=[gr.Image(label="Detected Hand"), gr.Label(num_top_classes=5, label="Predicted Digit")], live=True, )