Hi, one question, ¿What is the minimum size of each image for the code in lesson 1?, i want to try with another dataset and i don’t know that, thanks.
I believe you can supply any size. The transforms will rightsize them.
If the images are too small then its possible that training/recognition may not be that good. But that is as expected.
thanks @sanjeev.b, when you say “too small”, What do you mean?, do you thing that 299px by side is good?, thank you for you quick answer and your time.
Yes should be fine. resnet takes 224x224 (sz parameter in lesson 1 sheet).
I’ve used 64x64 no problem.