This repository was a fork of BVLC/caffe and includes the upsample, bn, dense_image_data and softmax_with_loss (with class weighting) layers of caffe-segnet (https://github.com/alexgkendall/caffe-segnet) to run SegNet with cuDNN version 5.
see above link.
I'm using caffe-segnet (the newest version). While do validation during training, I found that the test accuracy changes when I change batch size in TEST phase.
Issue summary
validation accuracy changes when batch_size value changes. opened in bvlc caffe. https://github.com/BVLC/caffe/issues/5621#issuecomment-368446563
Steps to reproduce
see above link. I'm using caffe-segnet (the newest version). While do validation during training, I found that the test accuracy changes when I change batch size in TEST phase.