rishikksh20/ceit-pytorch

Implementation of Convolutional enhanced image Transformer

ceit-pytorch's Issues

Changing in Image Size Problem

Hi @rishikksh20,

Thank you for your great work. When I give an image 224x224, it works fine. Then, I just changed image 224 to 112x112 and it gives an error. Any advice?

Code

"CeiT": CeiT(
        GPU_ID=GPU_ID,
        image_size=112,
        patch_size=4,
        dim=512,
        depth=20,
        num_classes=NUM_CLASS,
        heads=8,
        dropout=0.1,
        emb_dropout=0.1
        )

Traceback

    raise EinopsError(' Error while computing {!r}\n {}'.format(self, e))
einops.EinopsError:  Error while computing Rearrange('b c (h w) -> b c h w', h=14, w=14)
 Shape mismatch, 49 != 196

Image Size for CIFAR10 and CIFAR100

I wonder why the image size for cifar is 224?

Is this just for convenience?

请问下ReAttention得作用是啥呢

Question about parameters Initialization

Hello, I'm appreciate that your work help me a lot. I have a question that dose Transformer model need parameters initialization?

Error while training on CIFAR!!

raise EinopsError('Shape mismatch, {} != {}'.format(length, known_product))
einops.EinopsError: Shape mismatch, 4 != 196

Training on new datasets.

Are there any instructions to help me train the model on other datasets mentioned in the paper, e.g., Oxford pets?
As I understood, as of now, it is just possible to train on CIFAR10 or CIFAR100.
I would appreciate any clues that point me in the right direction.
Thanks a lot.