search

WongKinYiu / yolor

implementation of paper - You Only Learn One Representation: Unified Network for Multiple Tasks (https://arxiv.org/abs/2105.04206)
GNU General Public License v3.0
1.99k stars 518 forks source link

Training Yolor_p6 in darknet #157

Open Wazaki-Ou opened 2 years ago

Wazaki-Ou commented 2 years ago

I tried to find a way to do this without having to post an issue, but I couldn't get any hint so far. I really hope someone can help.

So I'm trying to use Alexey repository to train YOLOR_p6 on my custom dataset, but there hasn't been any update concerning yolor over there (the latest enhancement was regarding yolo_csp). So I tried on my own, and while the cfg file is available here, I can only find the weights as .pt files.

Is there a way to convert the .pt file into a weight file that I can use with the cfg file to train using darknet? (Since I assume darknet cannot use pytorch weights), or is it preferable to avoid training with darknet and just use pytorch to train?

If Pytorch is the best (or only) way, will I need to reorganize my dataset and labels (already organized to match darknet format) ?

Thanks a lot for any help or advice !!

Edit: It seems like even the cfg file needs to be converted to darknet format according to a post I came across. Can anyone please confirm that as well?

SanKumSan commented 2 years ago

@Wazaki-Ou I would like to do this approach as well, could you please share, what all you found the process. It would be a great resource. Thanks