Could you provide clean_train_filenames.pickle?

[ShunyuanZheng]

Could you please provide the clean train data list in segm/data/coco.py's L85, since there may exist something different with the one generated on the whole training set of ImageNet.

[ShunyuanZheng]

Thanks for your answering. But when run the training script, a .pickle file will be actually generated according to segm/data/coco.py's L88-L100 which covers all ImageNet training samples.

[shuchenweng]

Thanks for your answering. But when run the training script, a .pickle file will be actually generated according to segm/data/coco.py's L88-L100 which covers all ImageNet training samples.

Hi, sorry for the late reply, we provide the clean_train_filenames.pickle here, where we filter out the grayscale images in ImageNet without RGB ground truth. Unzip it and use it please.

[ShunyuanZheng]

Thanks for your answering. But when run the training script, a .pickle file will be actually generated according to segm/data/coco.py's L88-L100 which covers all ImageNet training samples.

Hi, sorry for the late reply, we provide the clean_train_filenames.pickle here, where we filter out the grayscale images in ImageNet without RGB ground truth. Unzip it and use it please.

Thanks!