MicroNet config for Human Pose Estimation task

[pha-nguyen]

Hi, thanks for your great work!

I am attempting to make this structure work on the pose estimation task. However, I found the modification in your paper which makes me confused:

The models are modified for the keypoint detection task, by increasing the resolution (×2) of a select set of blocks (all blocks with stride of 32). Each model contains a head with three micro-blocks (one of stride 8 and two of stride 4) and a pointwise convolution that generates heatmaps for 17 keypoints.

• increasing the resolution (×2): How can you do that? By decreasing the kernel size or manually editing which number?
• blocks with stride of 32, stride 8, stride 4: I could not find such any configurations. As I understood, stride is usually set to 1 or 2, and uncommonly 3 or more. Also in microconfig.py, your stride setting is no greater than 2.

It would be great for me if you could provide the configuration file and the modified implementation. Thanks in advance!

[liyunsheng13]

Hi, I appreciate your interest in our work. However, the code for key point detection is still under processing. We will release soon. you can refer to the code for all the implementation details.

[notplus]

Hi, I appreciate your interest in our work. However, the code for key point detection is still under processing. We will release soon. you can refer to the code for all the implementation details.

Looking forward to your code. :heart_eyes:

[10qiu01]

Hello, I am also looking for the code. Hahaha, how is it going?