EXIT

It may be something with the GPU environment. Have you tried with only 1 GPU and 2 GPUs? (```export CUDA_VISIBLE_DEVICES=0```)

Are you using clusters or multiprocess? The code uses [DataParallel](https://github.com/exitudio/GaitMixer/blob/main/src/train.py#L188) so it doesn't support multiprocess.

One difference from GaitGraph is we use Triplet loss from [pytorch_metric_learning ](https://github.com/KevinMusgrave/pytorch-metric-learning). But it shouldn't be a problem. It also works on my 4 GPU server.

You can try ```--loss_func supcon``` to see that the Triplet loss causes this problem or not.

Seem like this problem is from the inverse kinematics (and forward kinematics ) format of [text-to-motion](https://github.com/EricGuo5513/text-to-motion) is different from [smplpytorch](https://github.com/gulvarol/smplpytorch). However, I still can't solve the problem.

Hi, Thank you for being interested in our work. the variance we're seeing is likely because the CASIA-B dataset is relatively small, and the inclusion of bag (BG) and coat...

Yes. We use the preprocessed data from GaitGraph.

Hi, We use GradCAM to visualize where the model attend to. Since, we didn't clean up the visualization code so we remove it but you still can see our code...

I assume that you are trying to compare with the appearance base methods, right? The model-based (skeleton-based) methods are a lot smaller models because they use skeleton data instead of...

We didn't explore much about the pose estimator. We experimented with [SimCC](https://github.com/leeyegy/SimCC). But it's not a clear improvement so we just follow GaitGraph which uses HRNet.