Open nightrain-vampire opened 8 months ago
According to the error report,it seems that something is wrong with your model parameters. The checkpoint's image_size should be 224, and it's patch_size is 16, so the shape of pos_embed is [1, 197, 768]. But your init-model pos_embed is [1, 4097, 768].So I guess that your change the parameter "imgae_size" from 224 into 1024 with the "patch_size" remained 16.Maybe restoring the super-parameter would be a solution.
I'm also unable to get the weights to load :/
I try to use mae_vit_base in the demo, but it reports:
My code is below:
What's the matter with the pretrained model? I also tried 'mae_pretrain_vit_base_full.pth', but it reports the same error. Can anyone help?