Reviews: Deep Neural Networks with Box Convolutions

Thank you for the clarifications. I am inclined to keep my ratings. One concern I still have is the new approach (BoxERFNet) doesn't seem to be giving any gains in Cityscapes (test) so I not sure how effective the approach is. The results on the SUN RGBD are good but as mentioned, the earlier models were not tuned for that dataset, so it is hard to make a strong conclusion. Integral images used a trick for fast computation and has been used for many problems in computer vision but had not been used in neural networks yet.