search

CommissarMa / Context-Aware_Crowd_Counting-pytorch

The implementation of Context-Aware Crowd Counting(CVPR2019)
MIT License
68 stars 23 forks source link

Is batch_size=1 best? #12

Closed Becmamba closed 5 years ago

Becmamba commented 5 years ago

Since the amount of data is not so big, 300, can I set batch_size bigger?

CommissarMa commented 5 years ago

If you use ShanghaiTech A to train the model, the batch size must be 1 since the images have different resolution. If use ShanghaiTech B , you can set arbitrary batch_size. A bigger batch_size can accelerate convergence.