Image Classification on ImageNet (val) (Top-1 Acc, Hardware Latency, Efficiency)
0.792Top-1 AccuracyMathNAS
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| MathNASFLOPs (M)=669, Search Method=On-Device Search, Switch Unit=Block2023.11 | 0.792 | 2,253 | 1,398 | 81 | — | 61 | 3,125 | 781 | |
| Dynamic-OFAFLOPs (M)=732, Search Method=Random+Evolution, Switch Unit=Network2023.11 | 0.781 | 2,404 | 1,485 | 80 | 35 | 244 | 7 | 1.7 | |
| MathNASFLOPs (M)=257, Search Method=On-Device Search, Switch Unit=Block2023.11 | 0.759 | 832 | 525 | 76 | — | 61 | 3,125 | 781 | |
| AS-MNASNetFLOPs (M)=532, Search Method=Greedy Slimming, Switch Unit=Channel2023.11 | 0.754 | 2,097 | 1,457 | 207 | 4,000 | 37 | 4 | 1 | |
| US-MbNet-v2FLOPs (M)=300, Search Method=Manual Design, Switch Unit=Channel2023.11 | 0.715 | 1,358 | 959 | 158 | — | 18 | 27 | 6.7 | |
| S-MbNet-v2FLOPs (M)=301, Search Method=Manual Design, Switch Unit=Channel2023.11 | 0.705 | 1,346 | 958 | 118 | — | 15 | 4 | 1 |