search

backseason / PoolNet

Code for our CVPR 2019 paper "A Simple Pooling-Based Design for Real-Time Salient Object Detection"
MIT License
630 stars 153 forks source link

About dataloader #51

Closed GrassBro closed 4 years ago

GrassBro commented 4 years ago

Hi, @backseason ,

I utilize my own dataloader to train the network, where the image is resized and cropped to a fixed size. Therefore, I could use a large batch size, e.g., 32. I wonder to know whether such an operation would influence the final performance?

Thx!

backseason commented 4 years ago

There is a big chance that the final performance will be influenced. But I'm not sure whether it's in a positive way. Some of the hyper-parameters should be fine-tuned accordingly.