Image Classification on ImageNet (val) (Latency and Throughput)
793.1Throughput (Train Data Load) (FPS)JPEG-S
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| JPEG-SArchitecture=ViT-S, Color Space=DCT, Aug. Space=DCT, Mixed Precision=true2022.11 | 793.1 | — | 1.42 | 2.37 | 821.7 | 764 | 665.8 | 2,384.1 | — | 711.8 | 75.8 | |
| ResNet50Architecture=ResNet-50, Color Space=DCT, Aug. Space=RGB to DCT, Pipeline=Gueguen/Xu et al.2022.11 | 785.7 | — | — | — | — | — | — | 5,969.8 | — | 753.5 | 76.1 | |
| JPEG-SArchitecture=ViT-S, Color Space=DCT, Aug. Space=DCT2022.11 | 782.9 | — | 1.49 | 2.4 | 824.8 | 364.3 | 360.5 | 1,139.7 | — | 711.1 | 76.5 | |
| JPEG-TiArchitecture=ViT-Ti, Color Space=DCT, Aug. Space=DCT2022.11 | 775.3 | — | 1.42 | 2.62 | 816.2 | 857.2 | 687.6 | 2,847.5 | — | 752.3 | 75.1 | |
| JPEG-SArchitecture=ViT-S, Color Space=DCT, Aug. Space=RGB to DCT, Mixed Precision=true, Pipeline=Gueguen et al.2022.11 | 748.7 | — | 3.62 | 5.98 | 425.3 | 751.9 | 372.3 | 2,382.4 | — | 721 | 76.3 | |
| ResNet50Architecture=ResNet-50, Color Space=RGB, Aug. Space=RGB2022.11 | 688.2 | — | — | — | — | — | — | 1,226.8 | — | 639.1 | 76.1 | |
| ViT-SArchitecture=ViT-S, Color Space=RGB, Aug. Space=RGB, Mixed Precision=true2022.11 | 680.3 | — | 3.69 | 2.63 | 574.7 | 716.8 | 489 | 2,335.1 | — | 644.2 | 75.6 | |
| SwinV2-TArchitecture=SwinV2-T, Color Space=DCT, Aug. Space=DCT, Mixed Precision=true2022.11 | 664.2 | — | 1.4 | 2.47 | 752 | 241.9 | 235.9 | 824.9 | — | 578 | 79.4 | |
| ViT-SArchitecture=ViT-S, Color Space=RGB, Aug. Space=RGB2022.11 | 660.2 | — | 3.62 | 3.01 | 558 | 355.7 | 352.1 | 1,174.5 | — | 610.5 | 76.5 | |
| ViT-TiArchitecture=ViT-Ti, Color Space=RGB, Aug. Space=RGB2022.11 | 641.4 | — | 3.68 | 3.08 | 571.5 | 832.8 | 493.8 | 2,898.7 | — | 638 | 74.1 | |
| SwinV2-TArchitecture=SwinV2-T, Color Space=RGB, Aug. Space=RGB, Mixed Precision=true2022.11 | 614.8 | — | 3.61 | 2.98 | 489.8 | 231.5 | 231.6 | 809.5 | — | 516.2 | 79 |