src/predict.py

import cv2
import torch
import numpy as np
import pandas as pd
from .islr.islr_model import DummyISLRModel
from .pose.keypoints_utils import extract_keypoints_from_frame

#model = DummyISLRModel(num_classes=100)
#model.load_state_dict(torch.load("demo_model.pt", map_location='cpu'))
#model.eval()

LABELS = [f"Clase {i}" for i in range(100)]

def predict_from_video(video_path,model=None):
    cap = cv2.VideoCapture(video_path)
    keypoints = []

    while cap.isOpened():
        ret, frame = cap.read()
        if not ret:
            break
        keypoint = extract_keypoints_from_frame(frame)
        keypoints.append(keypoint)

    cap.release()
    if not keypoints:
        return "No keypoints detected", pd.DataFrame()

    x = torch.tensor(np.mean(keypoints, axis=0)).float().unsqueeze(0)
    #print("x:")
    #print(x)
    with torch.no_grad():
        logits = model(x)
        probs = torch.softmax(logits, dim=1).numpy()[0]

    #print("probs:")
    #print(probs)
    top5_idx = probs.argsort()[-5:][::-1]

    top5_labels = [LABELS[i] for i in top5_idx]
    top5_probs = [probs[i] for i in top5_idx]
    confidences = {LABELS[i]: float(probs[i]) for i in top5_idx}    

    #print("confidences:")
    #print(confidences)
    #df = pd.DataFrame({"label": top5_labels, "value": top5_probs})
    return top5_labels[0],confidences # df