Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Unable to test votenet pretrained model #145

Closed
filaPro opened this issue Oct 5, 2020 · 3 comments · Fixed by #163
Closed

Unable to test votenet pretrained model #145

filaPro opened this issue Oct 5, 2020 · 3 comments · Fixed by #163
Assignees
Labels
bug Something isn't working reimplementation

Comments

@filaPro
Copy link
Contributor

filaPro commented Oct 5, 2020

Looks like the state dict of current master votenet is not compatible with given pretrained checkpoints.

The model and loaded state dict do not match exactly

unexpected key in source state_dict: bbox_head.conv_pred.0.conv.weight, bbox_head.conv_pred.0.conv.bias, bbox_head.conv_pred.0.bn.weight, bbox_head.conv_pred.0.bn.bias, bbox_head.conv_pred.0.bn.running_mean, bbox_head.conv_pred.0.bn.running_var, bbox_head.conv_pred.0.bn.num_batches_tracked, bbox_head.conv_pred.1.conv.weight, bbox_head.conv_pred.1.conv.bias, bbox_head.conv_pred.1.bn.weight, bbox_head.conv_pred.1.bn.bias, bbox_head.conv_pred.1.bn.running_mean, bbox_head.conv_pred.1.bn.running_var, bbox_head.conv_pred.1.bn.num_batches_tracked, bbox_head.conv_pred.conv_out.weight, bbox_head.conv_pred.conv_out.bias

missing keys in source state_dict: bbox_head.conv_pred.shared_convs.layer0.conv.weight, bbox_head.conv_pred.shared_convs.layer0.bn.weight, bbox_head.conv_pred.shared_convs.layer0.bn.bias, bbox_head.conv_pred.shared_convs.layer0.bn.running_mean, bbox_head.conv_pred.shared_convs.layer0.bn.running_var, bbox_head.conv_pred.shared_convs.layer1.conv.weight, bbox_head.conv_pred.shared_convs.layer1.bn.weight, bbox_head.conv_pred.shared_convs.layer1.bn.bias, bbox_head.conv_pred.shared_convs.layer1.bn.running_mean, bbox_head.conv_pred.shared_convs.layer1.bn.running_var, bbox_head.conv_pred.conv_cls.weight, bbox_head.conv_pred.conv_cls.bias, bbox_head.conv_pred.conv_reg.weight, bbox_head.conv_pred.conv_reg.bias
@ZwwWayne
Copy link
Collaborator

ZwwWayne commented Oct 6, 2020

There is a refactor on the votehead and we forget to convert the keys in the checkpoints. @encore-zhou will have a check and update a model upgrade script. This might take some time, so for a quick fix, you might need to convert the key names by yourself.

@ZwwWayne ZwwWayne added the bug Something isn't working label Oct 6, 2020
@filaPro
Copy link
Contributor Author

filaPro commented Oct 6, 2020

There are 16 unexpected keys and 14 missing ones, so simple renaming is not going to help. Looking forward for update.

@filaPro
Copy link
Contributor Author

filaPro commented Oct 9, 2020

If it helps, votenet model for sunrgbd compatible with mmdetection==0.6.0: epoch_36.pth, 20201008_113526.log.json.
| AP_0.25 | AR_0.25 | AP_0.50 | AR_0.50 |
| 0.6015 | 0.8696 | 0.3598 | 0.5367 |

@ZwwWayne ZwwWayne linked a pull request Oct 19, 2020 that will close this issue
tpoisonooo pushed a commit to tpoisonooo/mmdetection3d that referenced this issue Sep 5, 2022
* add a demo for rewrite

* remove register symbolics

* place two image in one line

* width = 300

* use table to show two images

* fix typo
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug Something isn't working reimplementation
Projects
None yet
Development

Successfully merging a pull request may close this issue.

3 participants