Deep Isometric Learning for Visual Recognition
This work challenges the necessity of standard techniques in deep learning for visual recognition, potentially simplifying network architectures.
The paper tackles the problem of training deep convolutional neural networks without normalization or skip connections, achieving competitive performance on ImageNet and better than ResNet on COCO by enforcing near-isometric convolution kernels and using a modified ReLU.
Initialization, normalization, and skip connections are believed to be three indispensable techniques for training very deep convolutional neural networks and obtaining state-of-the-art performance. This paper shows that deep vanilla ConvNets without normalization nor skip connections can also be trained to achieve surprisingly good performance on standard image recognition benchmarks. This is achieved by enforcing the convolution kernels to be near isometric during initialization and training, as well as by using a variant of ReLU that is shifted towards being isometric. Further experiments show that if combined with skip connections, such near isometric networks can achieve performances on par with (for ImageNet) and better than (for COCO) the standard ResNet, even without normalization at all. Our code is available at https://github.com/HaozhiQi/ISONet.