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

Is there a pretrained weight file based on vanilla DETR? #21

Closed
PhanTask opened this issue Oct 2, 2021 · 4 comments
Closed

Is there a pretrained weight file based on vanilla DETR? #21

PhanTask opened this issue Oct 2, 2021 · 4 comments

Comments

@PhanTask
Copy link
Contributor

PhanTask commented Oct 2, 2021

The pretrained weight files of Trackformer currently are all based on Deformable DETR. However, my current environment does not support Deformable DETR due to some physical limitations.

Could you please provide a pretrained weight file based on vanilla DETR (e.g., DETR-ResNet-50)? Thanks!

@timmeinhardt
Copy link
Owner

Are you referring to a model file pretrained on CrowdHuman or COCO detection? The latter can be found in the original DETR repository.

@PhanTask
Copy link
Contributor Author

PhanTask commented Oct 5, 2021

Thank you, @timmeinhardt
I remember for vanilla-DETR-based TrackFormer, there are some additional attention layers (layers_track_attention) for projecting output embeddings to track queries. If I am not mistaken, the pretrained vanilla DETR does not have weights for these layers. Do you have pretrained ready-to-use weight files for vanilla-DETR-based TrackFormer?

@timmeinhardt
Copy link
Owner

timmeinhardt commented Oct 5, 2021

We provide the model file for our MOTS20 experiments which are still vanilla-DETR-based. These should include the weights for the additional tracking attention layers. You should be able discard the segmentation head and work with the rest for pure bounding box based tracking.

@PhanTask
Copy link
Contributor Author

PhanTask commented Oct 5, 2021

@timmeinhardt Great! Thanks a lot!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants