Does anyone have suggestions on a model that can do well on classifying hundreds or thousands of categories?

… Of black and white images.
Preferably with pre-trained weights.
I’ve been using Vgg with batchnorm. As a test run I have been classifying 85 categories, and that alone took around 9 hours. Assuming the time is going to linearly scale (because for each new category I’d be adding new data), for 850 categories I’d need days, let alone thousands.
Moreover, the vgg model was designed for color images with 3 channels. Hence using vgg for black/white images may have resulted in unnecessary training. I wonder if there’s a pre-trained model dealing with black and white.


Maybe you can just use one gray channel so that the computation time can be reduced?