search

ZF4444 / MMAL-Net

This is a PyTorch implementation of the paper "Multi-branch and Multi-scale Attention Learning for Fine-Grained Visual Categorization (MMAL-Net)" (Fan Zhang, Meng Li, Guisheng Zhai, Yizhao Liu).
249 stars 57 forks source link

Another N_list config [for speed up training] #4

Closed ahkarami closed 4 years ago

ahkarami commented 4 years ago

@ZF1044404254, Thank you very much for your great work. Can you please introduce another configuration for N_list (for CAR data set) that will increase the speed of training? (Do we need to change window_side & ratios?)

ZF4444 commented 4 years ago

@ahkarami Hi, of course, you can reduce N1, N2, N3 of N_list which can improve the training speed a little, but the accuracy may also be impaired. As for window_side is related to the depth of the network, so for the currently used backbone (resnet-50), window_side is 32; ratios is the aspect ratio of each window, you can try to increase or decrease the number of proportions (ratios), maybe make the accuracy there is improvement. N_list and ratios are hyperparameters, which can be determined through a lot of experiments.

ahkarami commented 4 years ago

Thank you very much for your useful comment.