search

thohemp / 6DRepNet360

Official Pytorch implementation of "Towards Robust and Unconstrained Full Range of Rotation Head Pose Estimation" IEEE TIP 24
MIT License
106 stars 2 forks source link

Fine-tuned model? #7

Open Redhwan-A opened 4 months ago

Redhwan-A commented 4 months ago

Hi,

Could you make your fine-tuned model available, as you did in 6DRepNet? please.

I would like to test my own datasets on your own fine-tuned model for the comparison for narrow and full-range datasets.

Thank you in advance!

thohemp commented 4 months ago

Hello,

you will find a link to the main model here: https://github.com/thohemp/6DRepNet360/blob/master/sixdrepnet360/test.py#L140 The other ones will be added soon.

Redhwan-A commented 4 months ago

@thohemp, thank you so much for your reply. This fine-tuned model for narrow range ("https://cloud.ovgu.de/s/2sP3yLrEwyfSmqC/download/6DRepNet360_300W_LP_AFLW2000.pth").

Really, I am interested in the fine-tuned model for the full range for CMU datasets.

Thanks.

thohemp commented 4 months ago

I think the notation of the file is misleading. I updated the readme with both models.

Redhwan-A commented 4 months ago

Thank you so much for your update. I ran your code on the CMU dataset and used your fine-tuned model. I got this result for CMU dataset for my CMU dataset:

Yaw: 7.0174, Pitch: 10.5485, Roll: 7.6579, MAE: 8.4079
Vec1: 11.5907, Vec2: 8.3935, Vec3: 11.6801, VMAE: 10.5548

360