stefanopini / simple-higherhrnet Goto Github PK

View Code? Open in Web Editor NEW

152.0 152.0 18.0 15.58 MB

Multi-person Human Pose Estimation with HigherHRNet in Pytorch, with TensorRT support

License: GNU General Public License v3.0

Python 75.44% Jupyter Notebook 24.56%

coco-dataset deep-learning higher-hrnet hrnet human-pose-estimation keypoint-detection mscoco-keypoint pytorch

simple-higherhrnet's People

Contributors

Stargazers

Watchers

Forkers

cyy0523xc souring001 baofengzan freegliboracle jianqiangren saccadic alexis-mignon wolfworld6 wukongxuetang zruiqian ajinkyapuar zardzen bstandaert abdur-rahim-sheikh kinqsradio anzisheng aruldeva273 gerald0juni0r

simple-higherhrnet's Issues

HRnet and HigherHRnet compare

Hi, thanks for sharing, I can understand the reasoning is slow due to the post processing complexity, but isn’t this model an updated version of HRNET? Previous approaches to detection and pose estimation have been used to infer key points for multiple people, and it is not clear why this is not as effective as HRNET

RuntimeError: [CamGear:ERROR] :: Source is invalid, CamGear failed to initialize stream on this source!

Hi, sorry for asking so many juvenile questions.
When I run !python scripts/live-demo.py --camera_id 0 in Googlecolaboratory, I get RuntimeError: [CamGear:ERROR] :: Source is invalid, CamGear failed to initialize stream on this source!
The webcam is recognized on GoogleCoraboratory.
I would be glad to know how to solve this problem.
Thank you in advance.

tensorrt

Did you try to convert this into tensorrt？

Can't use 48_640 model

 model = SimpleHigherHRNet(                                                                                                                                                                                                                                                                                                                                                                
      c=48,                                                                                                                                                                                                                                                                                                                                                                                 
      nof_joints=17,                                                                                                                                                                                                                                                                                                                                                                        
      checkpoint_path="./weights/pose_higher_hrnet_w48_640.pth",                                                                                                                                                                                                                                                                                                                            
      resolution=640,                                                                                                                                                                                                                                                                                                                                                                       
      device="cuda:0",

results in this

RuntimeError: Error(s) in loading state_dict for HigherHRNet:
	Missing key(s) in state_dict:

The 32_512 model works fine

torch.onnx.export come across an error :"Process finished with exit code -1073741819 (0xC0000005)"

When I run "export-tensorrt-model.py", it crush at calling "torch.onnx.export", the return info is "Process finished with exit code -1073741819 (0xC0000005)"
Anyone can tell me the reason of this bug?

When will the training code be released?

Thank for your great work.

Two group of predicted joints are nealy the same in a single image.

I print out the predicted joints:

[[[5.5107422e+02 5.4550781e+01 8.2366186e-01]
[5.4369141e+02 5.8769531e+01 8.2028019e-01]
[5.4158203e+02 5.6660156e+01 5.7210848e-02]
[5.4896484e+02 7.1425781e+01 8.4656513e-01]
[5.5212891e+02 6.8261719e+01 7.4948044e-03]
[5.8587891e+02 9.5683594e+01 7.1548438e-01]
[5.8482422e+02 6.9316406e+01 6.5927768e-01]
[6.3544922e+02 9.8847656e+01 6.0702914e-01]
[6.3123047e+02 5.9824219e+01 3.3245289e-01]
[6.6498047e+02 6.1933594e+01 5.5162466e-01]
[6.5126953e+02 4.7167969e+01 2.7785197e-01]
[6.7552734e+02 8.9355469e+01 6.8746221e-01]
[6.7130859e+02 7.0371094e+01 6.1395431e-01]
[7.3037109e+02 1.0306641e+02 5.1185238e-01]
[7.3458984e+02 5.0332031e+01 4.9360916e-01]
[7.7888672e+02 1.2837891e+02 6.7314917e-01]
[7.8205078e+02 2.7128906e+01 7.0142037e-01]]

[[5.5107422e+02 5.4550781e+01 8.2366186e-01]
[5.4369141e+02 5.8769531e+01 8.2028019e-01]
[5.4158203e+02 5.6660156e+01 5.7210848e-02]
[5.4896484e+02 7.1425781e+01 8.4656513e-01]
[5.5212891e+02 6.8261719e+01 7.4948044e-03]
[5.8587891e+02 9.5683594e+01 7.1548438e-01]
[5.7955078e+02 1.7705273e+03 3.0691934e-01]
[6.3544922e+02 9.8847656e+01 6.0702914e-01]
[6.3123047e+02 5.9824219e+01 3.3245289e-01]
[6.6498047e+02 6.1933594e+01 5.5162466e-01]
[6.5126953e+02 4.7167969e+01 2.7785197e-01]
[6.7552734e+02 8.9355469e+01 6.8746221e-01]
[6.7130859e+02 7.0371094e+01 6.1395431e-01]
[7.3037109e+02 1.0306641e+02 5.1185238e-01]
[7.3458984e+02 5.0332031e+01 4.9360916e-01]
[7.7888672e+02 1.2837891e+02 6.7314917e-01]
[7.8205078e+02 2.7128906e+01 7.0142037e-01]]]

As is shown above, two group of predicted joints are nealy the same except for the 6th row "right_shoulder".

I suppose something went wrong in the keypoint match part. How could I sove it?

Hi is there weight file available for MPII

I tried and it is working. How can I try with MPII key annotations? I think it needs a proper weights for MPII shaped pose estimation

Where will the video be saved?

Hi,Thanks for the great code.
I am running it on GoogleColaboratory.
When I run it with a saved video, it only saves a still image called 'frame.png'.
Where is the video with heatmap saved?
If you need to write any additional code, I would be glad to know.
Thank you in advance for your help.

Extract keypoints in csv file

Hi~
Thank you for the code implementation

Is live_demo.py has the function of generating keypoints in csv file?
Thank you!

How to train for multi-person detection?

I have a dataset which is with skeleton annotations.
However, there are multi-person. Not single person.
How can I train this?

I do not want to first detect the bounding box of the person.

Error when it isnt any person on image

Hello. Thank you for very useful model.
I get the error

expected a non-empty list of Tensors

when I try to predict pose on image without any person. When person is on it I have successfull estimation.

COCO skeleton connection

simple-HigherHRNet/misc/visualization.py

Line 38 in 59ece4a

[0, 5], [0, 6]

why do you use [0, 5], [0, 6] instead of [3, 5], [4, 6]？

Error loading pretrained model

Hello. Thank for your great work.

The pretrained model
pose_higher_hrnet_w32_512 works.
But the pose_higher_hrnet_w48_640.pth.tar
does not work. There is an error.

I have cloned the project after you made an change according to the closed issue.
How can I solve this?!

.csv file

Hello,What is the meaning of each parameter in the.csv file? If you have time, I hope you can answer this question.thank you