Designing Network Design Spaces

In this work, we present a new network design paradigm. Our goal is to help advance the understanding of network design and discover design principles that generalize across settings... Instead of focusing on designing individual network instances, we design network design spaces that parametrize populations of networks. The overall process is analogous to classic manual design of networks, but elevated to the design space level. Using our methodology we explore the structure aspect of network design and arrive at a low-dimensional design space consisting of simple, regular networks that we call RegNet. The core insight of the RegNet parametrization is surprisingly simple: widths and depths of good networks can be explained by a quantized linear function. We analyze the RegNet design space and arrive at interesting findings that do not match the current practice of network design. The RegNet design space provides simple and fast networks that work well across a wide range of flop regimes. Under comparable training settings and flops, the RegNet models outperform the popular EfficientNet models while being up to 5x faster on GPUs. read more

PDF Abstract CVPR 2020 PDF CVPR 2020 Abstract


Results from the Paper

Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Image Classification ImageNet RegNetY-800MF Top 1 Accuracy 76.3% # 426
Number of params 6.3M # 285
Image Classification ImageNet RegNetY-600MF Top 1 Accuracy 75.5% # 448
Number of params 6.1M # 286
Image Classification ImageNet RegNetY-4.0GF Top 1 Accuracy 79.4% # 326
Number of params 20.6M # 232
Image Classification ImageNet RegNetY-1.6GF Top 1 Accuracy 78.8% # 355
Number of params 11.2M # 258
Image Classification ImageNet RegNetY-8.0GF Top 1 Accuracy 79.9% # 311
Number of params 39.2M # 169
Hardware Burden None # 1
Operations per network pass None # 1