sparkfish/augraphy

stucked when I process image in the cycle

kukugpt opened this issue · 5 comments

kukugpt commented

ink_phase = [Letterpress(blur=0.0, n_samples=(100,300), n_clusters=(100,350), std_range=(1000, 3000), value_range=(180,255), p=p), # foreground
]
paper_phase = [DirtyScreen(p=p), #background
NoiseTexturize(sigma_range=(1,5), turbulence_range=(1,4), p=0.8), #background
]

post_phase = [ ]
pipeline = AugraphyPipeline(ink_phase, paper_phase, post_phase)

for img_path in tqdm(img_file_list[27:]):
img_path = img_path.strip()
img = cv2.imread(img_path)
aug_imgx2 = pipeline(img)

kwcckw commented

Hi, what do you mean by stuck here? Is the Python still running? Or it crashes?

kukugpt commented

python is running but can not obtain the result

kukugpt commented

This problem seems to be a problem in NoiseTexturize, when exec "augmentations/noisetexturize.py", line 74, in noise
result = np.random.normal(0, sigma, size=(ysize, xsize)"

kukugpt commented

it made by turbulence_range=(1,4), which I have made a pull request

  • 👍1
kwcckw commented

Thanks, i was able to identify the root cause and it's due to an infinity loop caused by ratio = (ratio // turbulence) or 1. So turbulence must > 1, else ratio will never == 1. I will push a patch shortly.