About the bottom up method

[CheungBH]

Hello. Are there any papers about the bottom-up method? Like the model structure, image preprocesses, and loss function.

[innerlee]

See the cites in docs here https://mmpose.readthedocs.io/en/latest/bottom_up_models.html#associative-embedding-ae-hrnet

[jin-s13]

Currently, MMPose supports

@inproceedings{newell2017associative,
  title={Associative embedding: End-to-end learning for joint detection and grouping},
  author={Newell, Alejandro and Huang, Zhiao and Deng, Jia},
  booktitle={Advances in neural information processing systems},
  pages={2277--2287},
  year={2017}
}

and

@inproceedings{cheng2020higherhrnet,
  title={HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Estimation},
  author={Cheng, Bowen and Xiao, Bin and Wang, Jingdong and Shi, Honghui and Huang, Thomas S and Zhang, Lei},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={5386--5395},
  year={2020}
}

Other popular bottom-up papers include:

1. Realtime Multi-person 2D Pose Estimation Using Part Affinity Fields - Cao, Z., Simon, T., Wei, S., & Sheikh, Y. (CVPR 2017)
2. PifPaf: Composite Fields for Human Pose Estimation - Sven Kreiss, Lorenzo Bertoni, Alexandre Alahi (CVPR 2019)
3. MultiPoseNet: Fast Multi-Person Pose Estimation using Pose Residual Network - Muhammed Kocabas, Salih Karagoz, Emre Akbas (ECCV 2018)
4. Mixture Dense Regression for Object Detection and Human Pose Estimation - Ali Varamesh, Tinne Tuytelaars (CVPR 2020)

[CheungBH]

Thank you for your reply. By the way, I wonder are there any references mentioning how to determine the kernel size of gaussian distribution? I found them most of them are fixed values.

[jin-s13]

Please check this. Rethinking the Heatmap Regression for Bottom-up Human Pose Estimation. Luo et al. ArXiv 2021 In this paper, the authors propose a scale-adaptive heatmap regression (SAHR) method, which adaptively adjusts the Gaussian kernel.

[CheungBH]

Thanks for your help!