Ahmadzei's picture
added 3 more tables for large emb model
5fa1a76
The pre-training objective is relatively simple:
by masking a large portion (75%) of the image patches, the model must reconstruct raw pixel values.