chengeyang / human-pose-estimation-benchmarking-and-action-recognition Goto Github PK

I checked your code inside src/mylib/data_preprocessing.py
The code calculates both the keypoints coordinates and some of the joint angles. But only the x,y coordinates of the keypoints are used for the training If i am correct?

hello,I want to ask that when I was training the model with my own dataset,the loss is always nan , while the one-hot matrix is correct.Oh,I used 'categorical_crossentropy' to calculate the loss...Why the results were always nan?? and the accuracy was 0.2086 ,which was very low

I want to make my model.
I'm using tf-pose. But how to make skeleton_raw.csv and skeleton_filtered.csv?
Can you tell me about it?
Thank you!

Hi all
I have already downloaded the tf-pose-estimation repo.
I see that the installation of the action recognition needs to compile with the tf repo.
How can i run the action recognition with both repo downloaded?
thanks in advance

When the input camera feed is placed at an angle or when some of the landmarks points are hidden it gives incorrect results.
Does your training data include angled images and points

when i run git clone https://github.com/ildoonet/tf-pose-estimation i got fatal: repository 'https://github.com/ildoonet/tf-pose-estimation/' not found

I can't access this URL . what shou i do?

Hello,

Thanks for sharing this great repo!
Can I run the code on CPU or is GPU a must?

thanks,

Hello, i run this program with jetson nano and i got tensorflow.python.framework.errors_impl.ResourceExhaustedError: OOM when allocating tensor of shape [128] and type float
[[{{node TfPoseEstimator/Mconv3_stage4_L2/biases}}]]
Please help me to fix this problem
Thank You

Hi @ChengeYang.
How did we make our own datasets ??

cuda10
cudnn7.6.1
tensorflow-gpu 1.14.0

I meet the error while running the code. And I have no idea how to solve it

(1) Unknown: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
[[{{node TfPoseEstimator/conv1_1/Conv2D}}]]
[[strided_slice_3-0-0-TransposeNCHWToNHWC-LayoutOptimizer/_9]]

Hello, Could you please share the steps of converting video data to skeleton data?

hello,when i load the model ,the problem ocurrs.such as:
ValueError: Unknown layer:layers
if you can help ,i will apreciate it.thank you

@ChengeYang @felixchenfy @bigfacebear I want to perform action recognition in alphapose, can you please let me know if this can be done without the normalizing of keypoints in [-1,1] range as in Openpose?