Image Classification on ImageNet (val) (Latency and Search Cost)
74.6Top-1 AccuracyProxyless-R (mobile)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Proxyless-R (mobile)Hardware-aware=true, No Proxy=true, No Repeat=true2018.12 | 74.6 | 92.2 | 78 | 200 | |
| AmoebaNet-AHardware-aware=false, No Proxy=false, No Repeat=false2018.12 | 74.5 | 92 | 190 | 75,600 | |
| Proxyless-G + LLHardware-aware=true, No Proxy=true, No Repeat=true2018.12 | 74.2 | 91.7 | 79 | 200 | |
| NASNet-AHardware-aware=false, No Proxy=false, No Repeat=false2018.12 | 74 | 91.3 | 183 | 48,000 | |
| MnasNetHardware-aware=false, No Proxy=false, No Repeat=false2018.12 | 74 | 91.8 | 76 | 40,000 | |
| MnasNet (our impl.)Hardware-aware=true, No Proxy=false, No Repeat=false2018.12 | 74 | 91.8 | 79 | 40,000 | |
| MobileNetV2No Repeat=false2018.12 | 72 | 91 | 75 | — | |
| Proxyless-G (mobile)Hardware-aware=true, No Proxy=true, No Repeat=false2018.12 | 71.8 | 90.3 | 83 | 200 | |
| MobileNetV1Hardware-aware=false, No Repeat=false2018.12 | 70.6 | 89.5 | 113 | — |