How to reconstruct some unlabeled images

facebookresearch / mae

PyTorch implementation of MAE https//arxiv.org/abs/2111.06377

Other

6.93k stars 1.17k forks source link

How to reconstruct some unlabeled images #170

Open young169 opened 11 months ago

young169 commented 11 months ago

the log file give me an erro like "Couldn't find any class folder in data/train." The images are not for classification, If I give all of them one fake label, will it work? To learn some knowledge from these data?

lennart-maack commented 11 months ago

As far as I understand, the labels are not used during pretraining. Maybe for finetuning afterwards. If you have a look into engine_pretrain.py line 39, they only use the samples and no labels

for data_iter_step, (samples, _) in enumerate(metric_logger.log_every(data_loader, print_freq, header)):