Tf-pose-estimation: save the estimated video

[guofuzheng]

i want to save the estimated video without the original background just the skeleton ,what should i do?

[AVCarreiro]

In src/estimator.py you have a class TFPoseEstimator with the method draw_humans which takes an image as input. Simply pass an empty image (all zeros) and you get an image with only the skeleton overlaid.

[guofuzheng]

@AVCarreiro hello,i want to get the output like this

and i modify the src/run_video.py like this

i replace the original input "image" with "test". But i get the wrong output ,may i ask for your help?

[AVCarreiro]

Seems like you have an issue with the image shape, or the dtype. I think opencv expects images as matrices of type 'uint8'.

[guofuzheng]

@AVCarreiro I have solved the problem.Thank you for your help.

[rawandeheliah]

@guofuzheng could you provide me please with code for saving the estimated video ?

[guofuzheng]

@rawandeheliah hello, i did not save the estimated video, i saved the estimated pictures. The code can process videos also can process pictures. About how to save the estimated pictures, it is very easy. Let me show you the code.

humans = e.inference(image)
image = TfPoseEstimator.draw_humans(image, humans, imgcopy=True)
# cv2.imshow('tf-pose-estimation result', image)
cv2.imwrite(path_save+tmp.split('.')[0]+'/'+str(i)+'.jpg',image)

'image' is the image you want to estimate. After the second line of the code, and if we use 'cv2.imshow' the code will show us the estimated image, if we use'cv2.imwrite' , the code will save the estimated image. If you have any problem about how to save the estimated image, just let me know.