Jas
@rubeea Please check the official [homepage](http://www.pri.kyoto-u.ac.jp/datasets/macaquepose/index.html).
It seems that you are using `dist_train.sh` to train the models. These processes are the DataLoader workers. It is highly recommended to use `slurm_train.sh` instead of `dist_train.sh`, even in single-machine...
Same questions.
Does anyone download the annotations successfully?
Can I just delete the `memory_format` ?
Thanks for your interest in our work. We plan to release the implementation of ZoomNet in [MMPose](https://github.com/open-mmlab/mmpose) soon. Are there any specific questions about the implementation?
The FeatureNet and the BodyNet together form a HRNet-w32 network. FeatureNet is the stem of HRNet, with {conv1+bn+conv2+bn} structure. And BodyNet is the rest of HRNet.
@YangYangGirl For FasterRCNN, we simply treat face, lefthand, and righthand as three different types of objects. Since the receptive field is large enough, we can get reasonable results.
1. 没有特别具体的规则，要求标注员脑补猜测大拇指和小拇指的位置。 2. 这个是模型预测的结果，而不是人工gt标注的结果。大拇指和小拇指不太容易区分，模型的精度不够高，没有区分出来。