A problem with pretrain dla model

[tuclen-3]

No description provided.

[tuclen-3]

Hi @Owen-Liuyuxuan ,
I have problem when i run dla model because link http://dl.yf.io/dla/models don't activate. I try with another pretrain in timm library but I have an error that "RuntimeError: Only one file(not dir) is allowed in the zipfile". What pretrain i can use to replace this dla-102 pretrain.?

Thank you for your help.

[idjoopal]

I have the same issue. Are there other paths to download pre-trained models?

[tuclen-3]

Hi @idjoopal,
In my opinion, this problem occurs because of conflict between PyTorch version. I tested another pretrain in this link https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/dla102-21f57b54.pth that it works in PyTorch with version is higher than 1.6.0. In my google collab is 1.12.1+cu113. But it can't work in PyTorch 1.4.0 version. Do you know how to convert its format? Thanks

[Owen-Liuyuxuan]

That is a good problem and the website is now offline.

I have reported the problem to the official repo first.

ucbdrive/dla#25

[tuclen-3]

Hi @Owen-Liuyuxuan,
The link is down. I downloaded pretrain in that to my local. Can you give how to read it with PyTorch version 1.4.0? I used torch.load() but it raise error: "RuntimeError: unexpected EOF, expected 381299 more bytes. The file might be corrupted." Thanks

[Owen-Liuyuxuan]

As reported here.

Now the link is back. You can download the official model to offline storage now.