Open ldfinfontainebleau opened 5 years ago
Can you show some example images that you have used?
Of course!!! That is a better one.
Have you calculate the ratio(size of object decetion machine's output image divide by size which is your training data ) ? It is just a thought, I think it maybe means something.
Is your training datasets open ? Or have you used flip function on your training data? Nowadays, I did some comparative experiments and the results show that points on left hand are easier to be found. Hope your responds. Thank u!
@ldfinfontainebleau @vikasguptaiisc i tried to run the handpose code but i run into this error "output = net.forward() cv2.error: /io/opencv/modules/dnn/src/dnn.cpp:1430: error: (-215) output_slice.isContinuous() && output_slice.size == curr_output.size in function fuseLayers" , when i did an survey on this error it was mentioned that we have give proper channels in the image based on the prototxt, but even i give the proper image i get this can you share some pointers on this ??
@abhigoku10 I got the same bug in opencv3.3
@logic1988 what did you do to solve it
Thank you for doing this !!! I used this handpose very well, but there is a question. In many cases, this model can not found most of points(like it only can found less 10 in 21, maybe it is the issue about threshold value). When I cut images into small pieces which is full of a bare hand, the model did not show a better performance as my excepted !!! So I have a few questions. Is background affect so much or not ? Hopefully, I pronounce it clearly.