HRNET with Skeletal heatmap

Introduction

This is an pytorch implementation HRNet with skeletal heatmap. I am interested in semantic information of feature maps with each resolution in HRNet. So, I make ground truth heatmap of joints(default), skeletal, upper/lower and full body. Then, train the model while each predicted feature maps are trained with aforementioned GT heatmaps. The algorithm used to create heatmap referred to Human Pose Estimation Using Skeletal Heatmaps . Implementation is only for COCO and HRNet (No ResNet & No MPII).

Results on COCO val2017 with detector having human AP of 56.4 on COCO val2017 dataset

Arch	Input size	#Params	GFLOPs	AP	Ap .5	AP .75	AP (M)	AP (L)	AR	AR .5	AR .75	AR (M)	AR (L)
pose_hrnet_w48	384x288	65.4M	33.3	pose_hrnet	0.710	0.915	0.793	0.683	0.757	0.741	0.923	0.811	0.708

Note:

• Flip test is used.
• Person detector has person AP of 56.4 on COCO val2017 dataset.
• GFLOPs is for convolution and linear layers only.

Visualization

Visualizing results on COCO training

Keypoint Heatmap

Skeletal Heatmap

Half body Heatmap

Full body Heatmap