bottom up

[alicera]

Is the bottom - up network similar to openpose ? Because I get the onnx model output is [1,34,128,128] for (512,512,3) image. Is the 34 compose of 17 keypoints and 17 paf?

Thanks

[jin-s13]

No. Currently we support AE for bottom-up pose estimation.

@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}
}

[alicera]

For more detailed usage and alternative for per parameter in each module, please refer to the API documentation.

Do you know where is the API documentation for buttom-up config?

I want to know the config

test_cfg=dict(
num_joints=channel_cfg['dataset_joints'],
max_num_people=30,
scale_factor=[1],
with_heatmaps=[True],
with_ae=[True],
project2image=True,
nms_kernel=5,
nms_padding=2,
tag_per_joint=True,
detection_threshold=0.1,
tag_threshold=1,
use_detection_val=True,
ignore_too_much=False,
adjust=True,
refine=True,
flip_test=False))

https://github.com/open-mmlab/mmpose/blob/master/docs/tutorials/0_config.md#config-system-for-top-down-human-pose-estimation

[alicera]

I cant understand the parameter project2image (bool): Option to resize to base scale.

[wuyuuu]

No. Currently we support AE for bottom-up pose estimation.

@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}
}

I also want to ask a related question.

So the first 17 channels are keypoint heatmaps and the last 17 channels are AE tags, right?

And by the way, for multi-stage higher resolution head, I see the output channel after refinement is 17 instead of 34. Does it mean that the refinement stage only refines heatmaps?

[jin-s13]

So the first 17 channels are keypoint heatmaps and the last 17 channels are AE tags, right? Yes.

And by the way, for multi-stage higher resolution head, I see the output channel after refinement is 17 instead of 34. Does it mean that the refinement stage only refines heatmaps?

That's for higher-resolution heatmaps.