search

JinluZhang1126 / MixSTE

Official implementation of CVPR 2022 paper(MixSTE: Seq2seq Mixed Spatio-Temporal Encoder for 3D Human Pose Estimation in Video)
192 stars 16 forks source link

Training log #15

Closed flyyyyer closed 1 year ago

flyyyyer commented 1 year ago

Could you provide the traning log and the pretrained model? It seems hard to convergence.

JinluZhang1126 commented 1 year ago

Can I ask you how many epochs you have trained?

JinluZhang1126 commented 1 year ago

Can I ask you how many epochs you have trained?

I can not attach the code anytime now, but the training log and checkpoint will be supplied as soon.

There are some info that may help you: I trained for about 30 hours in 2080Ti GPU to get convergence, and the training loss went down after about 2-4 epochs.