Can I request early access to the code? about transgeo2022 HOT 8 CLOSED

Sorry that it is behind schedule. I may not have enough time to clean it, but it should be released by this weekend.

from transgeo2022.

Thank you. Looking forward to it.

from transgeo2022.

The code is out. You may check if there is any bug. Currently it is not very clean, but should be runnable.

from transgeo2022.

Hi, Thanks for releasing the code. I am starting to set it up on my machine. I get the following error when loading deit from pre-trained weights. Is it a bug?

_Traceback (most recent call last):

File "/conda/lib/python3.8/site-packages/torch/multiprocessing/spawn.py", line 59, in wrap
fn(i, *args)
File "/workspace/TransGeo2022/train.py", line 218, in main_worker
model = TransGeo(args=args)
File "/workspace/TransGeo2022/model/TransGeo.py", line 43, in init
self.query_net = base_model(crop=False, img_size=self.size_grd, num_classes=args.dim)
File "/workspace/TransGeo2022/model/Deit.py", line 160, in deit_small_distilled_patch16_224
new_matrix = resize(matrix).permute(0, 2, 3, 1).reshape([1, -1, weight.shape[-1]])
File "/conda/lib/python3.8/site-packages/torchvision/transforms/transforms.py", line 195, in call
return F.resize(img, self.size, self.interpolation)
File "/conda/lib/python3.8/site-packages/torchvision/transforms/functional.py", line 229, in resize
raise TypeError('img should be PIL Image. Got {}'.format(type(img)))
TypeError: img should be PIL Image. Got <class 'torch.Tensor'>

Btw, I could run the code on 1 gpu by not using pretrained deit weights.

from transgeo2022.

What is the version of your torchvision and pytorch? Could you upgrade them and try again?

from transgeo2022.

https://pytorch.org/vision/main/generated/torchvision.transforms.Resize.html
The resize function should support both PIL and tensor as input, and it works for torchvision==0.11.1+cu113

from transgeo2022.

Thank you for your interest. Let me know if you are able to run it.

from transgeo2022.

Thanks. Upgrading torchvision helped. I am able to run the code. Appreciate the help!

from transgeo2022.