FBNet: Hardware-Aware Efficient ConvNet Design via Differentiable Neural Architecture Search

CVPR 2019 Bichen WuXiaoliang DaiPeizhao ZhangYanghan WangFei SunYiming WuYuandong TianPeter VajdaYangqing JiaKurt Keutzer

Designing accurate and efficient ConvNets for mobile devices is challenging because the design space is combinatorially large. Due to this, previous neural architecture search (NAS) methods are computationally expensive... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Image Classification ImageNet FBNet-C Top 1 Accuracy 74.9% # 127
Number of params 5.5M # 59

Methods used in the Paper