pre-training model

[jingziyou]

Hi, when the pre-training model will be released？

[AmeryXiong]

I will upload as soon as possible.

[DeTeam]

Would be great to be able to get the weights and play around locally with inference 🙏 🤩

[AmeryXiong]

I will update the most version code and weights in today or tomorrow. Sorry. i am little busy.

获取Outlook for Androidhttps://aka.ms/AAb9ysg

---

From: Timur Amirov @.> Sent: Thursday, July 27, 2023 8:35:29 PM To: AmeryXiong/MixDehazeNet @.> Cc: AmeryXiong @.>; Comment @.> Subject: Re: [AmeryXiong/MixDehazeNet] pre-training model (Issue #1)

Would be great to be able to get the weights and play around locally with inference 🙏 🤩

— Reply to this email directly, view it on GitHubhttps://github.com/AmeryXiong/MixDehazeNet/issues/1#issuecomment-1653534938, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AYNYMOGFXMIB3GMHTSKTQZ3XSJOBDANCNFSM6AAAAAAY2J7EF4. You are receiving this because you commented.Message ID: @.***>

[AmeryXiong]

I have updated !

[PINTO0309]

@AmeryXiong

Thank you!

It appears that read access rights have not been granted to the Google Drive URL in the pre-trained model. Can you grant read permission so that everyone can download the file?

[AmeryXiong]

hi, I have granted permission to everyone!

[PINTO0309]

@AmeryXiong Thank you! Hmmm...I still get a screen saying I don't have permissions when I click on the README link. https://drive.google.com/drive/folders/1ep6W4H3vNxshYjq71Tb3MzxrXGgaiM6C?usp=drive_link

Although the screen is in Japanese, it means Access rights required in English.

[AmeryXiong]

Maybe it is ok now! You can try again.

[PINTO0309]

It worked! Thanks! :+1:

[DeTeam]

Thanks @AmeryXiong! Tried model-s + outdoor, looks good!

For model-s and indoor, however, I get size mismatch for a whole bunch of blocks. Do you know what might be causing the issue?

Partial log sample:

RuntimeError: Error(s) in loading state_dict for MixDehazeNet:
    size mismatch for layer1.blocks.0.mlp.0.weight: copying a param with shape torch.Size([96, 24, 1, 1]) from checkpoint, the shape in current model is torch.Size([96, 72, 1, 1]).
    size mismatch for layer1.blocks.0.mlp2.0.weight: copying a param with shape torch.Size([96, 24, 1, 1]) from checkpoint, the shape in current model is torch.Size([96, 72, 1, 1]).
    size mismatch for layer1.blocks.1.mlp.0.weight: copying a param with shape torch.Size([96, 24, 1, 1]) from checkpoint, the shape in current model is torch.Size([96, 72, 1, 1]).
    size mismatch for layer1.blocks.1.mlp2.0.weight: copying a param with shape torch.Size([96, 24, 1, 1]) from checkpoint, the shape in current model is torch.Size([96, 72, 1, 1]).

[AmeryXiong]

model-s and indoor pretrain weight is the intermediate product of the experiment.