However, existing approaches are too expensive for case-by-case ConvNetĪrchitecture optimality depends on factors such as input resolution and targetĭevices. Due to this, previous neuralĪrchitecture search (NAS) methods are computationally expensive. The FBNet search takes 8 GPUs for only 27 hours, so the computational cost is only 216 GPU hours, or 421× faster than MnasNet, 222× faster than NASNet, 27.8× faster than PNASNet, and 1.33× faster than DARTS.Ĥ.2.Authors: Bichen Wu, Xiaoliang Dai, Peizhao Zhang, Yanghan Wang, Fei Sun, Yiming Wu, Yuandong Tian, Peter Vajda, Yangqing Jia, Kurt Keutzer Download PDF Abstract: Designing accurate and efficient ConvNets for mobile devices is challengingīecause the design space is combinatorially large.However, the search cost is orders of magnitude lower. Among all the automatically searched models, FBNet’s performance is much stronger than DARTS, PNASNet, and NASNet, and better than MnasNet.The FLOP count is 1.56×, 1.58×, and 1.03× smaller than MobileNetV2, ShuffleNet V2, and MnasNet-92.The latency is 28.1 ms, 1.33× and 1.19× faster than MobileNetV2 and ShuffleNet V2.
In the third group, FBNet-C achieves 74.9% accuracy, same as 2.0- ShuffleNet V2 and better than all others.Compared with MnasNet, FBNet-B’s accuracy is 0.1% higher, latency is 0.6ms lower, and FLOP count is 22M (relative 7%) smaller.In the second group, FBNet-B achieves comparable accuracy with 1.3- MobileNetV2, but the latency is 1.46× lower, and the FLOP count is 1.73× smaller, even smaller than 1.0- MobileNetV2 and 1.5- ShuffleNet V2.FBNet-A’s FLOP count is only 249M, 50M smaller (relative 20%) than MobileNetV2 and ShuffleNet V2, 20M (relative 8%) smaller than MnasNet, and 2.4× smaller than DARTS.Regarding latency, FBNet-A is 1.9 ms (relative 9.6%), 2.2 ms (relative 11%), and 8.6 ms (relative 43%) better than the MobileNetV2, ShuffleNet V2, and CondenseNet counterparts.In the first group, FBNet-A achieves 73.0% accuracy, better than 1.0- MobileNetV2 (+1.0%), 1.5- ShuffleNet V2 (+0.4%), and CondenseNet(+2%), and are on par with DARTS and MnasNet-65.The architecture distribution parameter θ is trained on the rest 20% of ImageNet training set with Adam. wa is trained on 80% of ImageNet training set using SGD with momentum.Samsung Galaxy S8 with a Qualcomm Snapdragon 835 platform is targeted.
0 kommentar(er)
