rom1504 / img2dataset

Easily turn large sets of image urls to an image dataset. Can download, resize and package 100M urls in 20h on one machine.
MIT License
3.74k stars 341 forks source link

Failed to resize 20% #440

Closed JosselinSomervilleRoberts closed 1 week ago

JosselinSomervilleRoberts commented 1 week ago

Hi, I am downloading datacomp.

I am getting about 20-25% failed to download which is expected I guess, but also 25% of failed to resize. Could you tell me what could cause a failure to resize (explanations and/or a pointer to the relevant code). We are using resize_mode="keep_ratio_largest" with min_image_size=200 and max_aspect_ratio=3.0

Thanks a lot for the help!

JosselinSomervilleRoberts commented 1 week ago

This was due to our image constraints (minimum size, max aspect, ...). Sorry for the inconvenience!