search

rabBit64 / Sign-language-recognition-with-RNN-and-Mediapipe

Apache License 2.0
9 stars 0 forks source link

Understand output text file #8

Open rami-alloush opened 4 years ago

rami-alloush commented 4 years ago

Hi, Thank you for your efforts. I've been running the Android example for the multihandtracking graph. On Android, we get 21 landmarks per hand, and for each hand we get (X,Y) and Z is optional if we speificy the 3d=true option. Can you explain the exact output in the text file from your script? Thanks

rami-alloush commented 4 years ago

I want to clarify that I understand that you just dump the x,y values to the text file as they arrive. My application is different and I want to know if this data was coming from only one hand in the frame or two hands (I was expecting 42 features at a time or 84 numbers (x ,y) per frame). Is that possible? Thanks

rabBit64 commented 4 years ago

Hello, the points you mentioned are correct and we extracted 84 numbers for both hands per frame.

rami-alloush commented 4 years ago

Thanks for the reply. I'm trying to separate the output from each frame into one line. I added out << "\n"; after the for loop below ofstream out(str, ios_base::out | ios_base::app); in the landmarks_to_render_data_calculator.cc file. The problem is that this separates each Hand not each Frame and it would work even if the frame has one hand. Is there a way to check the number of hands in the frame and output all the 84 features only if two hands present? (as in the Android Java example) Thanks and sorry for the long question.

sakshamprakash01 commented 4 years ago

Hello, I am not sure why but the text files only seem to contain 0s, 1s and -1s. Not the coordinates. Is there something I am missing? I am looking for the coordinates after feeding my videos.

rabBit64 commented 4 years ago

@sakshamprakash01 Hello, try updating landmarks_to_render_data_calculator.cc and build.py to the latest version. Here is the output description. Thank you.