Hello everyone,

I'm trying to execute the following command for getting the 3D pose output:

python src/openpose_3dpose_sandbox.py --camera_frame --residual --batch_norm --dropout 0.5 --max_norm --evaluateActionWise --use_sh --epochs 200 --load 4874200 --pose_estimation_json output/ --write_gif --gif_fps 24

But am getting the following error

experiments\All\dropout_0.5\epochs_200\lr_0.001\residual\depth_2\linear_size1024\batch_size_64\no_procrustes\maxnorm\batch_normalization\use_stacked_hourglass\predict_17
A subdirectory or file -p already exists.
Error occurred while processing: -p.
A subdirectory or file experiments\All\dropout_0.5\epochs_200\lr_0.001\residual\depth_2\linear_size1024\batch_size_64\no_procrustes\maxnorm\batch_normalization\use_stacked_hourglass\predict_17\log already exists.
Error occurred while processing: experiments\All\dropout_0.5\epochs_200\lr_0.001\residual\depth_2\linear_size1024\batch_size_64\no_procrustes\maxnorm\batch_normalization\use_stacked_hourglass\predict_17\log.
INFO:__main__:start reading json files
Traceback (most recent call last):
 File "src/openpose_3dpose_sandbox.py", line 415, in <module>
   tf.app.run()
 File "C:\Program Files\Anaconda3\envs\pose3Drnn\lib\site-packages\tensorflow\python\platform\app.py", line 125, in run
   _sys.exit(main(argv))
 File "src/openpose_3dpose_sandbox.py", line 232, in main
   smoothed = read_openpose_json()
 File "src/openpose_3dpose_sandbox.py", line 131, in read_openpose_json
   drop_curves_plot.savefig(pngName)
 File "C:\Program Files\Anaconda3\envs\pose3Drnn\lib\site-packages\matplotlib\pyplot.py", line 689, in savefig
   res = fig.savefig(*args, **kwargs)
 File "C:\Program Files\Anaconda3\envs\pose3Drnn\lib\site-packages\matplotlib\figure.py", line 2094, in savefig
   self.canvas.print_figure(fname, **kwargs)
 File "C:\Program Files\Anaconda3\envs\pose3Drnn\lib\site-packages\matplotlib\backend_bases.py", line 2075, in print_figure
   **kwargs)
 File "C:\Program Files\Anaconda3\envs\pose3Drnn\lib\site-packages\matplotlib\backends\backend_agg.py", line 521, in print_png
   cbook.open_file_cm(filename_or_obj, "wb") as fh:
 File "C:\Program Files\Anaconda3\envs\pose3Drnn\lib\contextlib.py", line 81, in __enter__
   return next(self.gen)
 File "C:\Program Files\Anaconda3\envs\pose3Drnn\lib\site-packages\matplotlib\cbook\__init__.py", line 407, in open_file_cm
fh, opened = to_filehandle(path_or_file, mode, True, encoding)
 File "C:\Program Files\Anaconda3\envs\pose3Drnn\lib\site-packages\matplotlib\cbook\__init__.py", line 392, in to_filehandle
   fh = open(fname, flag, encoding=encoding)
FileNotFoundError: [Errno 2] No such file or directory: 'gif_output/dirty_plot.png'

1. Operating system: Windows 10
2. TensorFlow version: 1.13.1
3. Python version:3.6.5

Can anyone help me out with this?

Thanks in advance

Hi! We're running the following code (we've done the 150 epochs) : python src/openpose_3dpose_sandbox.py --camera_frame --residual --batch_norm --dropout 0.5 --max_norm --evaluateActionWise --use_sh --epochs 150 --load 3655650 --openpose json_directory --write_gif --gif_fps 24

This is one of the output frames from Openpose:

This is the result we get from the code:

Operating system Windows, tensorflow v 1.12.0 , python v 3.6.6

Are we missing a flag? We tried with another video and we have the same issue with the legs

Thank you a lot.

Apparently HrNET , for 2d pose estimation has higher accuracy and i guess that will improve the lifting from the 2D keypoints much better. However the joint order is COCO keypoints order and thus the lifting fails. Is it possible to add it as an extension.

Hrnet Keypoints output 2d keypoints which returns 1x17x3 where 1 is single person , followed by 17 keypoints and confidnce scores. The output was modified a bit to match with the input required for json files in openpose but i think , there has to be some more modifications to be done ?

@una-dinosauria

1. operating system: win10
2. tensorflow version :1.4
3. python version :3.7 run python src/openpose_3dpose_sandbox.py --camera_frame --residual --batch_norm --dropout 0.5 --max_norm --evaluateActionWise --use_sh --epochs 200 --load 4874200 --pose_estimation_json /path/to/json_directory --write_gif --gif_fps 24 Used Pre-trained model .

Generated 3d_data.json and animation.gif 3d_data.zip animation.zip

but Why are there 32 points in 3d_data.json? How do these points correspond to joints?

Thanks for your interest in our research!

Hi, First off, thank you very much for sharing this great work! It has been incredibly helpful to have this available.

I have a quick question about OpenPose vs SH. I'm trying to estimate the pose based on the 16 feature point locations. I am using OpenPose for getting the estimation of pose in 2D. The features output by OpenPose can be found here: https://github.com/CMU-Perceptual-Computing-Lab/openpose/blob/master/doc/media/keypoints_pose.png

I noticed that stacked hourglass method outputs slightly different features and I am looking for ways to correct my data for that. The feature points that are missing/questionable in OpenPose are:

• hip center - not a huge problem, should be possible to just take the average from left/right hip
• spine - slightly more problematic as I don't know what sh means by "spine" and OpenPose does not provide a direct match
• thorax - could perhaps be replaced by base of neck?
• head - could be replaced by nose

My question is, have you tested or have you seen any results from using OpenPose and if you have, how were the results compared to sh after running it through your project(in your opinion, given that the model is trained on OP data)? Is it even worth correcting my data for the current pre-trained model or should I just go ahead and retrain using OP data/set up SH since the probability of getting any reasonable results otherwise would be very low?

I would love to have a minimal system running asap and am trying to find some ways to cut corners :)

Again, thank you very much for the great work that you have done. I hope the question isn't too much of a waste of time as this is not necessarily what the project was designed to do.

It's written in ReadMe

We are in the process of putting together a tutorial to prepare the data for this code. Please check back soon for an update.

when the tutorial to prepare the data is released?

Because I don't get a license for this dataset（human3.6）.I hope to get it from https://www.dropbox.com/s/e35qv3n6zlkouki/h36m.zip. Show remote server can't connect. I just want to get the 3D joint position. Can you give me a new URL. Thank you very much

Hi,

I am trying to run your code to infere 3d coordinates from the 2d coordinates from my own from one 2d keypoint extractor. I am converting my 2d positions into h5-files and storing them into S*/StackedHourglass/name.h5 but I am still having issues. It forces me to have 8 sequences, which I don't fully understand.

Could you please give me a hint? am I doing it correctly? i understand, storing my 2d coordinates into StackedHourglass folders and running "predict_3dpose.py" would store into folder "MyPoses" the predicted 3d coordinates from the input 2d coordinates, is it correct?

thanks in advance

Hello,

I noticed that each time a model.step() is performed, it needs the 3d data to be fed in as an input. Would we be able to perform inference without having the 3D ground truth for the keypoints?

Hi @una-dinosauria ,I am very appreciate for your job. when I read your code, a question puzzle for long time in the code

step_loss, loss_summary, poses3d = model.step( sess, enc_in, dec_out, dp, isTraining=False )

when getting the prediction pose3d,why I should the input of dec_out（the GT 3d） ?I try to reset the value dec_out to zero ,the return value of pose3d become different. Thx for your reply

Hi, Thanks for your wonderful works, but when I try the code in README for quick demo, there is no skeleton of 3dpose in visualization, and I found something strange at the end of training.

1. Operating system : windows 10

2. Tensorflow version : 2.3.0

3. Python version : 3.8.8

4. The stack trace of the error that you see

   ============================= Global step: 24371 Learning rate: 9.90e-04 Train loss avg: 0.1195 ============================= ===Action=== ==mm== Directions nan Discussion nan Eating nan Greeting nan Phoning nan Photo nan Posing nan Purchases nan Sitting nan SittingDown nan Smoking nan Waiting nan WalkDog nan Walking nan WalkTogether nan Average nan =================== Saving the model... done in 1979.55 ms

Hello,

Hope you are doing good.

I am trying to train the model for 1 epoch on my custom dataset , it trains but at last it shows Average : nan. Something is wrong right rather its should show some value.

This i am getting when traning 48 49 50 51 52 53 54 55 56] [ 0 1 2 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56] nan Average nan

Saving the model... done in 254.45 ms

Any idea what is the problem , where i should look at to fix this problem ?

Thanks for your interest in our research!

If you have problems running our code, please include

1. Your operating system
2. Your tensorflow version
3. Your python version
4. The stack trace of the error that you see

This is research code, and its primary purpose is to reproduce the scientific results that you see in our paper. If you are looking for ways to extend our work such as

• handling arbitrary images
• handling videos as input
• using your webcam as input
• using other 2d detectors
• large-scale deployment
• handling missing data
• you want to make a startup out of this code

then pull requests are welcome, but it is very unlikely we will have the time to help you.

Thanks for your research, but when I want to recurrent your results, the system always remind me of "Not found file, data/h36m/cameras.h5". I seek for the h36m dataset from classmate's clouddisk and he believes that's integral, but I couldn't found that file either when your download link has already failed. I'm very grateful if you can solve this question

I am not able to get data from the H3.6M official website and the data resources from the internet is not complete. I'd appreciate if you can help me.

I get the data from Internet but I only have code-v1.1. Maybe because of this, when I use your quick demo, the loss is nan. I'm sorry to disturb you, but can you help me solve the question?

Hello everyone,

I'm trying to rebuild the result of this code and follow the introduction that provide on this website, however, I got the wrong predicted joints, which are not the 16 joints we want to show the skeleton. Could anyone help me change the code to train the model with the joints we want?

1. Operating system : windows 10 home
2. Tensorflow version : 2.3.0
3. Python version : 3.8.8