Timezone: »

 
Poster
Deep Isometric Learning for Visual Recognition
Haozhi Qi · Chong You · Xiaolong Wang · Yi Ma · Jitendra Malik

Tue Jul 14 10:00 AM -- 10:45 AM & Tue Jul 14 11:00 PM -- 11:45 PM (PDT) @
in Poster Session 4 »

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.

Keywords: Architectures · Computer Vision · Optimization · Deep Learning - General

Author Information

Haozhi Qi (UC Berkeley)
Chong You (University of California, Berkeley)
Xiaolong Wang (UCSD)
Yi Ma (UC Berkeley)
Jitendra Malik (University of California at Berkeley)

More from the Same Authors