Open star4s opened 7 years ago
@star4s I just used the learning Rate setting in my github code. How does your curve look like? Is it similar to mine? (P.s. Different machines may have different random seeds. Every batch contains the images random sampled. So it may be a little different.)
This is the result for opts.train.learningRate = [1e-5ones(1,3) 1e-7ones(1,1)]; . What is the parameter 5 of ones(1,3) and the parameter 7 of ones(1,1)? You choose the parameter of 5 and 6. In this case, Super resolution did not work on my sample when I used the Mat-Data.
Every time trying tuning the parameter, opts.train.learningRate, I cannot make like your SRnet-v1-color-128.
This is your convergence graph for the result of Mat-data.
Before you modify last one program, this was thing which you did upload, the mat file. I did change the parameter opts.train.learningRate.
opts.train.learningRate = [1e-5ones(1,5) 1e-6ones(1,2)]; opts.train.learningRate = [1e-5ones(1,4) 1e-6ones(1,1)]; opts.train.learningRate = [1e-5ones(1,2) 1e-6ones(1,1)]; opts.train.learningRate = [1e-5ones(1,3) 1e-6ones(1,1)]; opts.train.learningRate = [1e-5ones(1,2) 1e-6ones(1,2)]; opts.train.learningRate = [1e-5ones(1,3) 1e-6ones(1,2)]; %opts.train.learningRate = [1e-5ones(1,5) 1e-6ones(1,1)]; %opts.train.learningRate = [1e-5ones(1,5) 1e-6ones(1,4)]; %opts.train.learningRate = [1e-5ones(1,6) 1e-6ones(1,1)]; %opts.train.learningRate = [1e-5ones(1,6) 1e-6ones(1,2)]; %opts.train.learningRate = [1e-3ones(1,1) 1e-3ones(1,1)]; %opts.train.learningRate = [1e-5ones(1,1) 1e-6ones(1,2)]; %opts.train.learningRate = [1e-6ones(1,3) 1e-7ones(1,1)]; %opts.train.learningRate = [1e-5ones(1,3) 1e-5ones(1,1)]; %opts.train.learningRate = [1e-5ones(1,3) 1e-7ones(1,1)]; %opts.train.learningRate = [1e-4ones(1,3) 1e-7ones(1,1)]; %opts.train.learningRate = [1e-3ones(1,8) 1e-8ones(1,3)]; %opts.train.learningRate = [1e-4ones(1,3) 1e-7ones(1,2)]; %opts.train.learningRate = [1e-4ones(1,5) 1e-7ones(1,1)];
Any one did not like your result. Do you remeber what the parameter was onopts.train.learningRate ?
thank you for your help, every time.