I just scraped the data from web to build a female, male and organization classifier. But it seems none of the techniques such as Cyclic learning rate, Data augmentation helps in improving model accuarcy. I am following steps as per lesson 1. Even I tried to adjust learning rate it doesn’t seem to help.
Here, is notebook
Your image size is 48 x 48 x 3 . Are all the images of same shape? I think it is better to give the size parameter as 48 and try to see if it works.
The models we’re using aren’t generally going to work great on images smaller than about 128x128, FYI. So maybe scrape some larger images?
Do you have any suggestions which model should we use when images are smaller than 128x128.
The popular CIFAR10 dataset is 32x32, so you could check the performance on that: https://github.com/kuangliu/pytorch-cifar