search

yingxin-jia / SuperGlue-pytorch

[SuperGlue: Learning Feature Matching with Graph Neural Networks] This repo includes PyTorch code for training the SuperGlue matching network on top of SIFT keypoints and descriptors.
Other
545 stars 126 forks source link

how can i ues the trianed pth to the origin supergluepretrain code #29

Open ZhouShunlong opened 1 year ago

Shuhul24 commented 1 year ago

Hey @ZhouShunlong did you get any way how to use the weights that we get by training the superglue with the pretrained model?

ZhouShunlong commented 1 year ago

It seems that we can't get any details about how superglue train it's model

---Original--- From: @.> Date: Tue, Nov 7, 2023 14:28 PM To: @.>; Cc: @.**@.>; Subject: Re: [yingxin-jia/SuperGlue-pytorch] how can i ues the trianed pth to the origin supergluepretrain code (Issue #29)

Hey @ZhouShunlong did you get any way how to use the weights that we get by training the superglue with the pretrained model?

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

Shuhul24 commented 1 year ago

Well what I observed is that there are some changes in the architecture that is being used in official SuperGlue such as in line 198 in models/superglue.py the default config are different than the one in the official one.

Also in line 249 (in models/superglue.py) the script is 

desc0 = desc0 + self.kenc(kpts0, torch.transpose(data['scores0'], 0, 1))
desc1 = desc1 + self.kenc(kpts1, torch.transpose(data['scores1'], 0, 1))

whereas in the official models/superglue.py it is

desc0 = desc0 + self.kenc(kpts0, data['scores0'])
desc1 = desc1 + self.kenc(kpts1, data['scores1'])

I tried importing the trained model weights in the official model but it wasn't working as there were plenty of errors coming around these lines.

ZhouShunlong commented 1 year ago

Can you speak in Chinese ？We can talk in WeChat with searching my phone number ：18800543847

---Original--- From: @.> Date: Wed, Nov 8, 2023 13:35 PM To: @.>; Cc: @.**@.>; Subject: Re: [yingxin-jia/SuperGlue-pytorch] how can i ues the trianed pth to the origin supergluepretrain code (Issue #29)

Well what I observed is that there are some changes in the architecture that is being used in official SuperGlue such as in line 198 in models/superglue.py the default config are different than the one in the official one.

Also in line 249 (in models/superglue.py) the script is desc0 = desc0 + self.kenc(kpts0, torch.transpose(data['scores0'], 0, 1)) desc1 = desc1 + self.kenc(kpts1, torch.transpose(data['scores1'], 0, 1))

whereas in the official models/superglue.py it is desc0 = desc0 + self.kenc(kpts0, data['scores0']) desc1 = desc1 + self.kenc(kpts1, data['scores1'])

I tried importing the trained model weights in the official model but it wasn't working as there were plenty of errors coming around these lines.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>