Qikai Li

> evaluation = dict(interval=1, metric='mAP', save_best='AP') Hi. I noticed that in the config file, you set the evaluation interval to be 1, which means the model will do evaluation every...

@GXYao7 Sorry for the late reply. Have you checked the losses during the training? Are they normal or abnormal?

> the data_ time is basically 0 This is normal because this indicated that the time for loading data samples is relatively small. > my loss always kept a low...

@Taylorminer Hi, thanks for using mmpose. This demo script `body3d_multiview_detect_and_regress_img_demo.py` only support CMU Panoptic data format currently, you can find more description about this demo script in [this doc](https://github.com/open-mmlab/mmpose/blob/master/demo/docs/3d_multiview_human_pose_demo.md). You...

@Camila641 Hi. Thanks for using mmpose. You can refer to [this config ](https://github.com/open-mmlab/mmpose/blob/master/configs/body/3d_kpt_mview_rgb_img/voxelpose/panoptic/voxelpose_prn64x64x64_cpn80x80x20_panoptic_cam5.py), which train VoxelPose on Panoptic dataset. You can modify the corresponding settings according to your own dataset.

> should the dataset in Panoptic format Yes. The images and camera calibration data should be formatted into the Panoptic style. > are 3D keypoint coordinates needed? Yes, the ground...

I think you may try [labelbee](https://github.com/open-mmlab/labelbee). Or you may read the detail about the [Panoptic dataset](http://domedb.perception.cs.cmu.edu/), and see if there are some guidances to get 3D keypoint coordinates annotations.

@ChenZhenGui Thanks for using MMPose 1.x. For now, we do not have the script to test inference speed in 1.x version. > You may try https://github.com/open-mmlab/mmpose/blob/master/tools/analysis/benchmark_inference.py You can try the...

> about the speed-test feature I saw in the mmdeploy project, Could you please share the llink to this feature? So that we may add it in mmpose as well.

你好。你可以参考这个 [demo](https://github.com/open-mmlab/mmpose/blob/master/demo/docs/3d_multiview_human_pose_demo.md)，关于相机参数这一块，代码处理的逻辑不一样，可能需要自己修改一下。 tools/test.py 会输出模型的预测结果，可能需要经过一定的后处理才能得到你想要的结果~