Less Wright

Hi @zsgj-Xxx, Thanks for opening the issue. We have not had a chance to test Ranger21 out on multi-gpu yet, but for sure some aspects of it need to be...

Hi @NoOneUST, I would recommend just starting with no modifications - i.e. just run Ranger as it's designed to try and establish smart defaults. After that, you can start adjusting,...

Hi @juntang-zhuang, In our paper, we only trained 60 epochs due to the cost of training. Typically people train 200 - 300 or so epochs on ImageNet if the goal...

Hi @wangcn99 - thanks for the update re: CBAM. I suspect there is more linkage between the DETR head and the backbone than initially thought. Thus, swapping in new backbones...

Hi @qbx2 , Thanks for the feedback! I think adding a manual sync is a good idea, along with a manual clear weights (for restarting new training). I will try...

Hi @neuronflow, Thanks for opening the issue! Ranger21 does maintain a basic state dict but for sure we need to update it with some additional data to ensure a clean...

Hi @tazu786 - I had no issue simply training the whole thing rather than doing the freezing of the head. However, it's a valid point that we should test freezing...

Hi @e-sha - thanks for pointing this out! Offhand, yes, it looks like 0 would be a better result but will need to test and see. Can you test it...

Hi @akashpalrecha - sure, thanks for letting me know. I'll make it in the next 2 hours here and post.

Here you go: [](https://zenodo.org/badge/latestdoi/204534960) or html: Hope that helps and please let me know if you cite it (and thanks for the push to get setup on Zenodo)