Index mismatch hazard in dataset_preprocessing/ffhq/crop_images_in_the_wild.py

[Bahri Batuhan Bilecen]

There is no assertion for len(img_files)==len(lm_files). If unsatisfied ( len(lm_files)<len(img_files), i.e. the face reconstruction backbone could not find landmarks for all given images ) this will eventually cause a hazard in the loop beginning on line 36, causing a continuous mismatch between images and their landmark files:

eg3d/dataset_preprocessing/ffhq/crop_images_in_the_wild.py

Lines 26 to 38 in 7cf1fd1

 lm_dir = os.path.join(args.indir, "detections") 
 img_files = sorted([x for x in os.listdir(args.indir) if x.lower().endswith(".png") or x.lower().endswith(".jpg")]) 
 lm_files = sorted([x for x in os.listdir(lm_dir) if x.endswith(".txt")]) 
  
 lm3d_std = load_lm3d("Deep3DFaceRecon_pytorch/BFM/")  
  
 out_dir = os.path.join(args.indir, "crop") 
 if not os.path.exists(out_dir): 
     os.makedirs(out_dir, exist_ok=True) 
      
 for img_file, lm_file in zip(img_files, lm_files): 
  
     img_path = os.path.join(args.indir, img_file)