Image Classification on ImageNet (val) (Top-1/Top-5 Accuracy, Throughput)
85.43Top-1 AccuracyE-SpikeFormer + A2SG
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| E-SpikeFormer + A2SGParam (M)=173.0, Power (mJ)=35.64, Time Steps=1 × 42026.05 | 85.43 | — | — | |
| ViT+JumboSize=D=768, J=3, Throughput=3.1K, Resolution=224x2242025.02 | 84.9 | 97.5 | — | |
| CMT-L# Params=74.7M, Resolution=288^2, # FLOPs=19.5B, Ratio=1x2021.07 | 84.8 | 97.1 | 150.4 | |
| E-SpikeFormerParam (M)=173.0, Power (mJ)=35.6, Time Steps=1 × 42026.05 | 84.7 | — | — | |
| BaselineBit width=FP, Backbone Network=ViT-B, Target Architecture=-, Venue=-2026.06 | 84.53 | — | — | |
| CMT-B# Params=45.7M, Resolution=256^2, # FLOPs=9.3B, Ratio=1x2021.07 | 84.5 | 96.9 | 285.4 | |
| EfficientNet-B7# Params=66M, Resolution=600^2, # FLOPs=37B, Ratio=1.9x2021.07 | 84.3 | 97 | 55.1 | |
| ViT+RegistersSize=D=768, R=16, Throughput=2.9K, Resolution=224x2242025.02 | 84.3 | 97.2 | — | |
| MaskAQBit width=4w8a, Backbone Network=ViT-B, Target Architecture=ViT, Venue=-2026.06 | 84.3 | — | — | |
| MimiQBit width=4w8a, Backbone Network=ViT-B, Target Architecture=ViT, Venue=AAAI’ 252026.06 | 84.2 | — | — | |
| PSAQ-ViT V2Bit width=4w8a, Backbone Network=ViT-B, Target Architecture=ViT, Venue=T-NNLS’ 232026.06 | 84.02 | — | — | |
| EfficientNet-B6# Params=43M, Resolution=528^2, # FLOPs=19.2B, Ratio=2.0x2021.07 | 84 | 96.8 | 96.9 | |
| EfficientNetV2-S# Params=22M, Resolution=224^2, # FLOPs=8.8B, Ratio=1x2021.07 | 83.9 | — | — | |
| CMT-S# Params=25.1M, Resolution=224^2, # FLOPs=4.0B, Ratio=1x2021.07 | 83.5 | 96.6 | 562.5 | |
| BoTNet-S1-128# Params=75.1M, Resolution=256^2, # FLOPs=19.3B, Ratio=2.1x2021.07 | 83.5 | 96.5 | — | |
| CVT-21_{384}# Params=31.5M, Resolution=384^2, # FLOPs=24.9B, Ratio=2.7x2021.07 | 83.3 | — | — | |
| Swin-B# Params=88M, Resolution=224^2, # FLOPs=15.4B, Ratio=1.5x2021.07 | 83.3 | — | 278.1 | |
| Twins-SVT-L# Params=99.2M, Resolution=224^2, # FLOPs=14.8B, Ratio=1.7x2021.07 | 83.3 | — | 288 | |
| CeiT-S_{384}# Params=24.2M, Resolution=384^2, # FLOPs=12.9B, Ratio=1.4x2021.07 | 83.3 | 96.5 | — | |
| Twins-SVT-B# Params=56.0M, Resolution=224^2, # FLOPs=8.3B, Ratio=2.1x2021.07 | 83.1 | — | — | |
| DeiT-B_{384}# Params=85.8M, Resolution=384^2, # FLOPs=55.6B, Ratio=6.0x2021.07 | 83.1 | — | 85.9 | |
| EfficientNet-B4# Params=19M, Resolution=380^2, # FLOPs=4.2B, Ratio=1x2021.07 | 82.9 | 96.4 | 349.4 | |
| TNT-B# Params=65.6M, Resolution=224^2, # FLOPs=14.1B, Ratio=1.5x2021.07 | 82.8 | 96.3 | — | |
| ViT+JumboSize=D=384, J=6, Throughput=7.6K, Resolution=224x2242025.02 | 82.6 | 96.5 | — | |
| AdaDFQBit width=4w8a, Backbone Network=ViT-B, Target Architecture=CNN, Venue=CVPR’ 232026.06 | 82.43 | — | — | |
| CeiT-S# Params=24.2M, Resolution=224^2, # FLOPs=4.5B, Ratio=1.1x2021.07 | 82 | 95.9 | — | |
| ViT+RegistersSize=D=384, R=16, Throughput=6.6K, Resolution=224x2242025.02 | 81.9 | 96 | — | |
| BaselineBit width=FP, Backbone Network=DeiT-B, Target Architecture=-, Venue=-2026.06 | 81.85 | — | — | |
| CMT-XS# Params=15.2M, Resolution=192^2, # FLOPs=1.5B, Ratio=1x2021.07 | 81.8 | 95.8 | 857.4 | |
| RegNetY-8GF# Params=39.2M, Resolution=224^2, # FLOPs=8.0B, Ratio=2x2021.07 | 81.7 | — | 591.6 | |
| GDFQBit width=4w8a, Backbone Network=ViT-B, Target Architecture=CNN, Venue=ECCV’ 202026.06 | 81.68 | — | — | |
| EfficientNet-B3# Params=12M, Resolution=300^2, # FLOPs=1.8B, Ratio=1.2x2021.07 | 81.6 | 95.7 | 732.1 | |
| CPVT-S-GAP# Params=23M, Resolution=224^2, # FLOPs=4.6B, Ratio=1.2x2021.07 | 81.5 | — | — | |
| MaskAQBit width=4w8a, Backbone Network=DeiT-B, Target Architecture=ViT, Venue=-2026.06 | 81.4 | — | — | |
| BaselineBit width=FP, Backbone Network=Swin-T, Target Architecture=-, Venue=-2026.06 | 81.35 | — | — | |
| MimiQBit width=4w8a, Backbone Network=DeiT-B, Target Architecture=ViT, Venue=AAAI’ 252026.06 | 81.34 | — | — | |
| Swin-T# Params=29M, Resolution=224^2, # FLOPs=4.5B, Ratio=1.1x2021.07 | 81.3 | — | 755.2 | |
| T2T-ViT-19# Params=39.0M, Resolution=224^2, # FLOPs=8.0B, Ratio=2x2021.07 | 81.2 | — | — | |
| PVT-M# Params=44.2M, Resolution=224^2, # FLOPs=6.7B, Ratio=1.7x2021.07 | 81.2 | — | 528.1 | |
| PSAQ-ViT V2Bit width=4w8a, Backbone Network=DeiT-B, Target Architecture=ViT, Venue=T-NNLS’ 232026.06 | 81.15 | — | — | |
| ACD-U2026.03 | 81 | 95.5 | — | |
| GKL-KDTeacher model=BEiT-Large, Student model=ResNet-50, Training scheme=strong training scheme2025.03 | 80.98 | — | — | |
| ResNeXt-101-64x4d# Params=84M, Resolution=224^2, # FLOPs=32B, Ratio=8x2021.07 | 80.9 | 95.6 | — | |
| KDTeacher model=BEiT-Large, Student model=ResNet-50, Training scheme=strong training scheme2025.03 | 80.89 | — | — | |
| DKDTeacher model=BEiT-Large, Student model=ResNet-50, Training scheme=strong training scheme2025.03 | 80.77 | — | — | |
| DISTTeacher model=BEiT-Large, Student model=ResNet-50, Training scheme=strong training scheme2025.03 | 80.7 | — | — | |
| AdaDFQBit width=4w8a, Backbone Network=DeiT-B, Target Architecture=CNN, Venue=CVPR’ 232026.06 | 80.49 | — | — | |
| MaskAQBit width=4w8a, Backbone Network=Swin-T, Target Architecture=ViT, Venue=-2026.06 | 80.19 | — | — | |
| MimiQBit width=4w8a, Backbone Network=Swin-T, Target Architecture=ViT, Venue=AAAI’ 252026.06 | 80.06 | — | — | |
| PSAQ-ViTBit width=4w8a, Backbone Network=DeiT-B, Target Architecture=ViT, Venue=ECCV’ 222026.06 | 80.05 | — | — | |
| GDFQBit width=4w8a, Backbone Network=DeiT-B, Target Architecture=CNN, Venue=ECCV’ 202026.06 | 80.03 | — | — | |
| PSAQ-ViT V2Bit width=4w8a, Backbone Network=Swin-T, Target Architecture=ViT, Venue=T-NNLS’ 232026.06 | 79.98 | — | — | |
| BaselineBit width=FP, Backbone Network=DeiT-S, Target Architecture=-, Venue=-2026.06 | 79.85 | — | — | |
| DeiT-S# Params=22M, Resolution=224^2, # FLOPs=4.6B, Ratio=3.1x2021.07 | 79.8 | — | 940.4 | |
| AdaDFQBit width=4w8a, Backbone Network=Swin-T, Target Architecture=CNN, Venue=CVPR’ 232026.06 | 79.7 | — | — | |
| MaskAQBit width=4w8a, Backbone Network=DeiT-S, Target Architecture=ViT, Venue=-2026.06 | 79.44 | — | — | |
| EfficientNet-B1# Params=7.8M, Resolution=240^2, # FLOPs=0.7B, Ratio=1.2x2021.07 | 79.1 | 94.4 | — | |
| CMT-Ti# Params=9.5M, Resolution=160^2, # FLOPs=0.6B, Ratio=1x2021.07 | 79.1 | 94.5 | 1,323.5 | |
| PSAQ-ViTBit width=4w8a, Backbone Network=Swin-T, Target Architecture=ViT, Venue=ECCV’ 222026.06 | 79.06 | — | — | |
| CoaT-Lite Mini# Params=11M, Resolution=224^2, # FLOPs=2.0B, Ratio=1.3x2021.07 | 78.9 | — | — | |
| GDFQBit width=4w8a, Backbone Network=Swin-T, Target Architecture=CNN, Venue=ECCV’ 202026.06 | 78.9 | — | — | |
| MaskAQBit width=5w5a, Backbone Network=DeiT-B, Target Architecture=ViT, Venue=-2026.06 | 78.89 | — | — | |
| Semi-RML++2026.03 | 78.7 | — | — | |
| E-SpikeFormer + A2SGParam (M)=10.0, Power (mJ)=2.78, Time Steps=1 × 42026.05 | 78.61 | — | — | |
| MaskAQBit width=5w5a, Backbone Network=ViT-B, Target Architecture=ViT, Venue=-2026.06 | 78.6 | — | — | |
| E-SpikeFormerParam (M)=10.0, Power (mJ)=3.0, Time Steps=1 × 42026.05 | 78.5 | — | — | |
| MimiQBit width=4w8a, Backbone Network=DeiT-S, Target Architecture=ViT, Venue=AAAI’ 252026.06 | 78.48 | — | — | |
| ViT+JumboSize=D=192, J=6, Throughput=20.4K, Resolution=224x2242025.02 | 78.4 | 94 | — | |
| PSAQ-ViT V2Bit width=4w8a, Backbone Network=DeiT-S, Target Architecture=ViT, Venue=T-NNLS’ 232026.06 | 78.27 | — | — | |
| MimiQBit width=5w5a, Backbone Network=DeiT-B, Target Architecture=ViT, Venue=AAAI’ 252026.06 | 78.2 | — | — | |
| MobileNetV4Size=hybrid-medium, Throughput=8.8K, Resolution=224x2242025.02 | 78.1 | 94.3 | — | |
| MimiQBit width=5w5a, Backbone Network=ViT-B, Target Architecture=ViT, Venue=AAAI’ 252026.06 | 78.09 | — | — | |
| ViT-B/16_{384}# Params=85.8M, Resolution=384^2, # FLOPs=77.9B, Ratio=8.4x2021.07 | 77.9 | — | 85.9 | |
| CLIPCleaner2026.03 | 77.8 | 92.1 | — | |
| RankMatch2026.03 | 77.4 | 94.3 | — | |
| Meta-SpikeFormerParam (M)=31.3, Power (mJ)=32.8, Time Steps=4 × 12026.05 | 77.2 | — | — | |
| NoiseBox+SS-KNN2026.03 | 77 | 91 | — | |
| AdaDFQBit width=4w8a, Backbone Network=DeiT-S, Target Architecture=CNN, Venue=CVPR’ 232026.06 | 76.92 | — | — | |
| EfficientViTSize=B1, Throughput=9.8K, Resolution=224x2242025.02 | 76.9 | 93.5 | — | |
| SV-Learner2026.03 | 76.9 | 93.9 | — | |
| MaskAQBit width=5w5a, Backbone Network=Swin-T, Target Architecture=ViT, Venue=-2026.06 | 76.82 | — | — | |
| PSAQ-ViTBit width=4w8a, Backbone Network=DeiT-S, Target Architecture=ViT, Venue=ECCV’ 222026.06 | 76.56 | — | — | |
| GDFQBit width=4w8a, Backbone Network=DeiT-S, Target Architecture=CNN, Venue=ECCV’ 202026.06 | 76.49 | — | — | |
| CLAMP-ViTBit width=4w8a, Backbone Network=DeiT-S, Target Architecture=ViT, Venue=ECCV’ 242026.06 | 76.44 | — | — | |
| MimiQBit width=5w5a, Backbone Network=Swin-T, Target Architecture=ViT, Venue=AAAI’ 252026.06 | 76.39 | — | — | |
| DenseNet-169# Params=14M, Resolution=224^2, # FLOPs=3.5B, Ratio=7x2021.07 | 76.2 | 93.2 | — | |
| ResNet-50# Params=25.6M, Resolution=224^2, # FLOPs=4.1B, Ratio=2.7x2021.07 | 76.2 | 92.9 | — | |
| Teacher (ResNet50)Distillation Manner=Teacher, Architecture=ResNet502025.03 | 76.16 | — | — | |
| GenQBit width=4w4a, Backbone Network=DeiT-B, Target Architecture=ViT, Venue=ECCV’ 242026.06 | 76.1 | — | — | |
| LSL2026.03 | 75.8 | 91.8 | — | |
| DivideMix2026.03 | 75.2 | 90.8 | — | |
| RigLBackbone=ResNet-50, Density=20%2025.09 | 75.1 | — | — | |
| CPVT-Ti-GAP# Params=6M, Resolution=224^2, # FLOPs=1.3B, Ratio=2.6x2021.07 | 74.9 | — | — | |
| MobileNetV4Size=conv-medium, Throughput=11.2K, Resolution=224x2242025.02 | 74.9 | 92.6 | — | |
| SpikformerParam (M)=66.3, Power (mJ)=21.5, Time Steps=4 × 12026.05 | 74.8 | — | — | |
| ViT+JumboSize=D=128, J=6, Throughput=31.3K, Resolution=224x2242025.02 | 74.6 | 91.9 | — | |
| ViT+RegistersSize=D=192, R=16, Throughput=16.7K, Resolution=224x2242025.02 | 74.5 | 92.3 | — | |
| MaskAQBit width=4w4a, Backbone Network=DeiT-B, Target Architecture=ViT, Venue=-2026.06 | 74.41 | — | — | |
| SHViTSize=S3, Throughput=23.7K, Resolution=224x2242025.02 | 74.3 | 92 | — | |
| PSAQ-ViT V2Bit width=5w5a, Backbone Network=ViT-B, Target Architecture=ViT, Venue=T-NNLS’ 232026.06 | 74.29 | — | — |