Image Classification on ImageNet 1K (Top-1 Accuracy, Memory, Search Time)
86.4Top-1 AccuracySDTv3
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| SDTv3Backbone Architecture=SDTv3, #Params.=173, Precision=32, Timestep=82026.06 | 86.4 | 692 | — | — | |
| AQ4SViT-BeamBackbone Architecture=SDTv3, #Params.=173, Precision=mixed, Timestep=82026.06 | 86.22 | 85 | 87.7 | 4,001 | |
| AQ4SViT-GreedyBackbone Architecture=SDTv3, #Params.=173, Precision=mixed, Timestep=82026.06 | 85.45 | 121 | 82.5 | 127 | |
| QSDTBackbone Architecture=SDTv2, #Params.=6.8, Precision=4, Timestep=42026.06 | 80.3 | 3.2 | 98.5 | — | |
| SDTv2Backbone Architecture=SDTv2, #Params.=55.4, Precision=32, Timestep=42026.06 | 78.9 | 211 | — | — | |
| AQ4SViT-GreedyBackbone Architecture=SDTv2, #Params.=55.4, Precision=mixed, Timestep=42026.06 | 77.96 | 63 | 70.1 | 39 | |
| AQ4SViT-BeamBackbone Architecture=SDTv2, #Params.=55.4, Precision=mixed, Timestep=42026.06 | 77.54 | 21 | 90 | 1,153 | |
| QSViTBackbone Architecture=SDTv2, #Params.=55.4, Precision=mixed, Timestep=42026.06 | 76.8 | 163 | 22.8 | 257 | |
| SpikformerBackbone Architecture=Spikformer, #Params.=66.3, Precision=32, Timestep=42026.06 | 74.81 | 253 | — | — | |
| SDTBackbone Architecture=SDT, #Params.=66.3, Precision=32, Timestep=42026.06 | 74.06 | 253 | — | — | |
| AQ4SViT-BeamBackbone Architecture=SDT, #Params.=66.3, Precision=mixed, Timestep=42026.06 | 73.9 | 33 | 87 | 2,853 | |
| AQ4SViT-GreedyBackbone Architecture=SDT, #Params.=66.3, Precision=mixed, Timestep=42026.06 | 73.2 | 77 | 69.6 | 98 |