classification model
lightweight
- tiny-fire model: top-1 accuracy-49.974% -> 51.10%
- squeezenet
- mobilenet v1
- peleenet
- IGCV 1
- Nasnet
- shufflenet v1
- mobilenet v2
- IGCV 2
- shufflenet v2
- IGCV 3
- mobilenet v3
- xception
heavy
- VGG
- resnet
- densenet
- highway network
- SE-Net
- Inception v3, v4
- EfficientNet
- Hrnet
Table
| network | top-1 accuracy | params(million) | flops/Madds(million) | latency(ms)-hardware |
|---|---|---|---|---|
| fire(tiny ssd) | 48.8 | |||
| alexnet | 57.2 | 60 | 720 | |
| squeezenet v1 | 57.5 | 1.25 | 1700 | |
| mobilenet v2 0.35 | 60.8 | 1.6 | 59.2 | |
| squeezenet v1-variant | 61 | |||
| shufflenetv2 0.5 | 60.3 | 41 | ||
| mobilenetv1 0.5 | 63.7 | 1.3 | 149 | |
| mnasnet | 64.9 | 1.9 | 65.1 | |
| mobilenetv3-small 0.75 | 65.4 | 2.4 | 44 | |
| mobilenetv1 160x160 | 67.2 | 4.2 | 290 | |
| mobilenetv3-small 1.0 | 67.4 | 2.9 | 66 | |
| mobilenetv1 0.75 | 68.4 | 2.6 | 325 | |
| mobilenetv1 192x192 | 69.1 | 4.2 | 418 | |
| shufflenetv2 | 69.4 | 146 | ||
| googlenet | 69.8 | 6.8 | 1550 | |
| mobilenetv1 | 70.6 | 4.24 | 575 | |
| peleenet | 71.3 | 2.8 | 508 | |
| VGG16 | 71.5 | 138 | 15346 | |
| shufflenetv1 1.5x | 71.5 | 3.4 | 292 | |
| mobilenetv2 | 72 | 3.4 | 300 | P1-78 P2-90 P3-61.3 |
| mobilenetv3-large 0.75 | 73.3 | 4 | 155 | |
| shufflenetv1 2x(g=3) | 73.7 | 5.2 | 524 | |
| inception v2 | 73.9 | |||
| Nasnet-A | 74 | 5.3 | 564 | |
| shufflenetv1 | 74.8 | 2300 | ||
| shufflenetv2 2x | 74.9 | 591 | ||
| densenet-41-variant | 75.0 | 1.13 | 545 | |
| mobilenetv3-large | 75.2 | 5.4 | 219 | P1-66 P2-77 P3-52.6 |
| mnasnet-A1 | 75.2 | 3.9 | 315 | |
| resnet50 | 76 | 26 | 4100 | |
| densenet169 | 76.2 | 14 | 3500 | |
| efficientnetB0 | 76.3 | 5.3 | 390 | |
| resnet152 | 77.8 | 60 | 11000 | |
| densenet121 | 78.65 | 6.87 | 2833 | |
| efficientnetB1 | 78.8 | 7.8 | 700 | |
| inception v3 | 78.8 | 24 | 5700 | |
| xception | 79 | 23 | 8400 | |
| efficientnetB2 | 79.8 | 9.2 | 1000 | |
| inception v4 | 80 | 48 | 13000 | |
| inception-resnet-v2 | 80.1 | 56 | 13000 | |
| resnet101 | 80.1 | |||
| resnext101 | 80.9 | 84 | 32000 | |
| efficientnetB3 | 81.1 | 12 | 1800 | |
| polynet | 81.3 | 92 | 35000 | |
| SEnet-154 | 81.3 | 20700 | ||
| efficientnetB4 | 82.6 | 19 | 4200 | |
| SEnet | 82.7 | 146 | 42000 | |
| Nasnet-A | 82.7 | 89 | 24000 | |
| PNasnet | 82.9 | 86 | 23000 | |
| efficientnetB5 | 83.3 | 30 | 9900 | |
| efficientnetB6 | 84 | 43 | 19000 | |
| Gpipe | 84.3 | 557 | ||
| efficientnetB7 | 84.4 | 66 | 37000 |