Body(2D,Kpt,Sview,Vid)¶

Posetrack18 Dataset¶

Posewarper + Hrnet + Posetrack18 on Posetrack18¶

PoseWarper (NeurIPS'2019)

@inproceedings{NIPS2019_gberta,
title = {Learning Temporal Pose Estimation from Sparsely Labeled Videos},
author = {Bertasius, Gedas and Feichtenhofer, Christoph, and Tran, Du and Shi, Jianbo, and Torresani, Lorenzo},
booktitle = {Advances in Neural Information Processing Systems 33},
year = {2019},
}

HRNet (CVPR'2019)

@inproceedings{sun2019deep,
  title={Deep high-resolution representation learning for human pose estimation},
  author={Sun, Ke and Xiao, Bin and Liu, Dong and Wang, Jingdong},
  booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
  pages={5693--5703},
  year={2019}
}

PoseTrack18 (CVPR'2018)

@inproceedings{andriluka2018posetrack,
  title={Posetrack: A benchmark for human pose estimation and tracking},
  author={Andriluka, Mykhaylo and Iqbal, Umar and Insafutdinov, Eldar and Pishchulin, Leonid and Milan, Anton and Gall, Juergen and Schiele, Bernt},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={5167--5176},
  year={2018}
}

COCO (ECCV'2014)

@inproceedings{lin2014microsoft,
  title={Microsoft coco: Common objects in context},
  author={Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Doll{\'a}r, Piotr and Zitnick, C Lawrence},
  booktitle={European conference on computer vision},
  pages={740--755},
  year={2014},
  organization={Springer}
}

Note that the training of PoseWarper can be split into two stages.

The first-stage is trained with the pre-trained checkpoint on COCO dataset, and the main backbone is fine-tuned on PoseTrack18 in a single-frame setting.

The second-stage is trained with the last checkpoint from the first stage, and the warping offsets are learned in a multi-frame setting while the backbone is frozen.

Results on PoseTrack2018 val with ground-truth bounding boxes

Arch	Input Size	Head	Shou	Elb	Wri	Hip	Knee	Ankl	Total	ckpt	log
pose_hrnet_w48	384x288	88.2	90.3	86.1	81.6	81.8	83.8	81.5	85.0	ckpt	log

Results on PoseTrack2018 val with precomputed human bounding boxes from PoseWarper supplementary data files from this link¹.

Arch	Input Size	Head	Shou	Elb	Wri	Hip	Knee	Ankl	Total	ckpt	log
pose_hrnet_w48	384x288	81.8	85.6	82.7	77.2	76.8	79.0	74.4	79.8	ckpt	log

¹ Please download the precomputed human bounding boxes on PoseTrack2018 val from $PoseWarper_supp_files/posetrack18_precomputed_boxes/val_boxes.json and place it here: $mmpose/data/posetrack18/posetrack18_precomputed_boxes/val_boxes.json to be consistent with the config. Please refer to DATA Preparation for more detail about data preparation.