Results with 3DPW train
raulTrial opened this issue · comments
Hi, Thanks for the nice project. Is there by any chance you have results/model trained on 3DPW too?
Hi, thanks for your interest! I haven't used 3DPW for training. I think that kind of protocol is now discouraged by the 3DPW authors and the whole dataset should be used for testing.
Hi, yes that makes complete sense. The whole HPE is riddled with models trained on different datasets and different evaluation techniques(for ex: full 3dpw test shows better results than only test set); so its a bit hard to gauge what actually makes a method work. But nonetheless you have a very nice approach.
I have some new, better models trained on even more datasets. The corresponding paper is already accepted for WACV'23, I'm finalizing the camera ready version.
I can share any type of eval results you need. As there are so many different datasets, eval protocols, eval measures, subsets (like actions), model configurations (backbone, striding, test-time augmentation) etc. it's hard to provide all the numbers, so that page here on Github is just a selection...
Congratulations!!! I'll wait for your paper.