Follow the line of sight to paper recently, want to use end-to-end model processing of deep learning, have mastered the other network models, but do not know from where, recently crazy opencv supplement knowledge, eyes ROI area is input the coordinates of corresponding points I have labeled good, each see a point cut a figure, after the interception of a large number of samples and then go to training, but training model, direct input output a graph, we can predict the coordinates of the point, but online video teaching, and books are mostly classified sample, I think this is a regression problems, there are some don't know how to do, we give some advice,

CodePudding user response:

Yes, you this is a regression problem, and you can this problem as a problem of feature point detection, first of all should make sure your data set is large enough, then keep the network loss of computer is correct, personally think that the problem is not complicated, should be able to get a good result,,

CodePudding user response:

This problem, you can mark first, and then directly training model

CodePudding user response:

I see online have directly to collect their own images as training data set is regression model but feel insecure because of distance what posture change will affect the result that data set how much