seekFire

> Thanks for your works! In Deep residual network, resnet34 use basicblock and resnet50/101 use bottleneck for practice performance. So what about hrnet? If hrnet replace basicblock by bottleneck in...

@hyz-xmaster I got it! Thank you very much!

@bowenc0221 How is the value(0.5) determined? Is it because the size of **final_heatmaps** is 0.5X that of original image ?

@Dantju I think I found the code mistake that caused the bad result, see [here](https://github.com/HRNet/HigherHRNet-Human-Pose-Estimation/issues/17#issuecomment-785770361).

@JiaweiMorris I also met recently

I use the coordinate value predicted by JointFormer directly, only make **z**-value of all points negative.

@Yogurt2019 Thank you very much! According to the your answer, I found the meaning of **keypoints[:, 2]**: you can download the **UCF-QNRF_ECCV18** dataset which used in this project & paper,...

So how to obtain models in **experimental_models**? Only model_25 can be available?

@davidpagnon Thank you very much! Yes, I know the performance of Openpose is no longer advanced, so I wonder whether all openpose model are bottom-up? Including wholebody135 and body+foot25B?

@davidpagnon Yes, the link is OK! Thank you very much! Because my job focuses on deployment rather than training, I usually find the trained weight file directly. If all models...