Hang Zhang

@bryanyzhu @chongruo Any thoughts on 3D ResNeSt?

Looks like we have 3rd party implementation for ResNeSt3D. Feel free to check it out: https://github.com/QiaoranC/tf_ResNeSt_RegNet_model

👍 will merge after uploading the weights.

> @Jerryzcn > > Have you been able to run ResNeST on mobile? What is the accuracy you are getting for 14 & 26? The results are reported https://gluon-cv.mxnet.io/model_zoo/classification.html#resnest Model...

The batch size shouldn't be only 1 during the training. If you're evaluating the model, please set the model to eval mode: `model.eval()`

If you only train with batch size = 1, I would suggest using `model.eval()` instead of `model.train()` and reduce the learning rate.

GluonCV has ResNeSt-26, which may be similar to what you need. https://cv.gluon.ai/model_zoo/classification.html#resnest

The easiest way to train a keypoint detector with ResNeSt may be using d2 wrapper https://github.com/zhanghang1989/ResNeSt/tree/master/d2 I would recommend try using that.

detectron2 is built upon pytorch . The implementation is also similar to torchvision one.

sorry that I am not familiar with onnx.