| import torch |
| import argparse |
| from pathlib import Path |
|
|
| import cv2 |
| from loguru import logger |
| from tqdm.auto import tqdm |
|
|
| from easy_dwpose import DWposeDetector |
|
|
| if __name__ == "__main__": |
| parser = argparse.ArgumentParser() |
| parser.add_argument("--input", type=Path, default="assets/") |
| parser.add_argument("--output_path", type=Path, default="results/") |
| args = parser.parse_args() |
|
|
| device = "cuda:0" if torch.cuda.is_available() else "cpu" |
| detector = DWposeDetector(device=device) |
| logger.info(f"Loaded detector") |
|
|
| logger.info(f"Starting inference on folder {args.input}") |
| files = sorted(args.input.iterdir()) |
|
|
| images = [file for file in files if file.suffix in [".jpg", ".jpeg", ".png"]] |
| logger.info(f"Found {len(images)} images") |
|
|
| args.output_path.mkdir(parents=True, exist_ok=True) |
|
|
| for image_path in tqdm(images): |
| image = cv2.imread(str(image_path)) |
| image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) |
|
|
| result = detector(image, output_type="np", include_hands=True, include_face=True) |
| result = cv2.cvtColor(result, cv2.COLOR_RGB2BGR) |
|
|
| output_path = str(image_path).replace(str(args.input), str(args.output_path)) |
| cv2.imwrite(output_path, result) |
|
|
| logger.info("Done") |
|
|
