Image Classification on N-MNIST (test)
99.7AccuracyGET
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GETInput Type=Token, Params (M)=4.52023.10 | 99.7 | — | |
| PLIFInput Type=Spike, Params (M)=4.7 - 17.12023.10 | 99.6 | — | |
| VMV-GCNInput Type=Voxel, Params (M)=0.92023.10 | 99.5 | — | |
| DT-SNNInput Type=Spike2023.10 | 99.5 | — | |
| EV-VGCNNPre-training=Without pretraining2021.06 | 99.4 | — | |
| EV-VGCNN (Ours)2021.06 | 99.4 | — | |
| EV-VGCNNInput Type=Voxel, Params (M)=0.82023.10 | 99.4 | — | |
| Wu et al. (2018)Bio-plausible=Partial2026.02 | 99.4 | — | |
| MVF-NetPre-training=Pretrained on ImageNet [9], Backbone=ResNet-34 [21]2021.06 | 99.3 | — | |
| Swin-T v2Input Type=Token, Params (M)=6.92023.10 | 99.3 | — | |
| Nested-TInput Type=Token, Params (M)=4.22023.10 | 99.3 | — | |
| EV-VGCNN (Ours)Variation=w/ SFRL2021.06 | 99.2 | — | |
| HATSBackbone/Classifier=linear SVM2019.06 | 99.1 | — | |
| ESTPre-training=Pretrained on ImageNet [9], Backbone=ResNet-34 [21]2021.06 | 99.1 | — | |
| HATS2021.06 | 99.1 | — | |
| HATSInput Type=Frame2023.10 | 99.1 | — | |
| EvSInput Type=Graph2023.10 | 99.1 | — | |
| ESTPre-training=Without pretraining, Backbone=ResNet-34 [21]2021.06 | 99 | — | |
| RG-CNNs2021.06 | 99 | — | |
| RG-CNNsInput Type=Voxel, Params (M)=19.52023.10 | 99 | — | |
| M-LSTMPre-training=Pretrained on ImageNet [9], Backbone=ResNet-34 [21]2021.06 | 98.9 | — | |
| SLAYERLayers=3, Learning Method=Backpropagation2021.10 | 98.89 | — | |
| STBPLayers=2, Learning Method=Spatial and temporal backpropagation2021.10 | 98.78 | — | |
| BPSNNLayers=3, Learning Method=Backpropagation2021.10 | 98.74 | — | |
| M-LSTMPre-training=Without pretraining, Backbone=ResNet-34 [21]2021.06 | 98.6 | — | |
| ANN-SNN (2021)Bio-plausible=Partial2026.02 | 98.5 | — | |
| Deep STDP2025.11 | 98.4 | — | |
| E2VIDEvaluation Protocol=fine-tuned, Backbone/Classifier=simple CNN2019.06 | 98.3 | — | |
| AMAEInput Type=Frame, Params (M)=21.82023.10 | 98.3 | — | |
| Sa-SNN2025.11 | 98.3 | — | |
| ST3D (ours)Spikes=~3800, Epochs=162025.11 | 98.17 | — | |
| MVF-NetPre-training=Without pretraining, Backbone=ResNet-34 [21]2021.06 | 98.1 | — | |
| MVF-NetInput Type=Frame, Params (M)=33.52023.10 | 98.1 | — | |
| 2D ClusterSpikes=~4200, Epochs=322025.11 | 97.8 | — | |
| UltraDPLIFtype=Spatial (Diffusion PDE), T=302026.02 | 97.68 | — | |
| DSpike+T=302026.02 | 97.65 | — | |
| PLIFT=302026.02 | 97.61 | — | |
| DSpikeT=302026.02 | 97.59 | — | |
| TTFS (baseline)Spikes=~5000, Epochs=322025.11 | 97.58 | — | |
| DSpike+T=102026.02 | 97.55 | — | |
| PLIFT=102026.02 | 97.53 | — | |
| NALSM8000Layers=2, Learning Method=astro-STDP, GD on last layer2021.10 | 97.51 | — | |
| FullPLIFT=102026.02 | 97.5 | — | |
| Ours (M5: Full Hybrid)Bio-plausible=Yes2026.02 | 97.49 | — | |
| LIFT=102026.02 | 97.48 | — | |
| FullPLIFT=302026.02 | 97.48 | — | |
| UltraDLIFtype=Spatial (Diffusion PDE), T=302026.02 | 97.46 | — | |
| Ours (M4: SNN+HGRN)Bio-plausible=Yes2026.02 | 97.44 | — | |
| DSpikeT=102026.02 | 97.39 | — | |
| AdaLIFT=102026.02 | 97.38 | — | |
| UltraDLIFtype=Spatial (Diffusion PDE), T=102026.02 | 97.38 | — | |
| UltraDPLIFtype=Spatial (Diffusion PDE), T=102026.02 | 97.38 | — | |
| AdaLIFT=302026.02 | 97.3 | — | |
| LIFT=302026.02 | 97.29 | — | |
| UltraPLIFtype=Temporal (LIF ODE), T=102026.02 | 96.33 | — | |
| AER-SNNLayers=2, Learning Method=Backpropagation2021.10 | 96.3 | — | |
| NALSM1000Layers=2, Learning Method=astro-STDP, GD on last layer2021.10 | 96.13 | — | |
| UltraLIFtype=Temporal (LIF ODE), T=102026.02 | 96.1 | — | |
| DECOLLELayers=3, Learning Method=Backpropagation2021.10 | 96 | — | |
| UltraLIFtype=Temporal (LIF ODE), T=302026.02 | 95.87 | — | |
| UltraPLIFtype=Temporal (LIF ODE), T=302026.02 | 95.77 | — | |
| PointNet++Input representation=Proposed representation2021.06 | 95.5 | — | |
| Diehl & Cook (2015)Bio-plausible=Yes (STDP)2026.02 | 95 | — | |
| UltraDLIFtype=Spatial (Diffusion PDE), T=12026.02 | 94.14 | — | |
| UltraDPLIFtype=Spatial (Diffusion PDE), T=12026.02 | 94.14 | — | |
| UltraPLIFtype=Temporal (LIF ODE), T=12026.02 | 93.11 | — | |
| SLAYER (2018)Bio-plausible=Yes2026.02 | 92.5 | — | |
| UltraLIFtype=Temporal (LIF ODE), T=12026.02 | 90.41 | — | |
| DSpikeT=12026.02 | 90.23 | — | |
| DSpike+T=12026.02 | 90.23 | — | |
| FullPLIFT=12026.02 | 89 | — | |
| LIFT=12026.02 | 88.54 | — | |
| PLIFT=12026.02 | 88.54 | — | |
| AdaLIFT=12026.02 | 88.54 | — | |
| PointNet++Input representation=Point-wise2021.06 | 84.1 | — | |
| HOTS2019.06 | 80.8 | — | |
| HOTS2021.06 | 80.8 | — | |
| E2VIDEvaluation Protocol=transfer learning2019.06 | 80.7 | — | |
| EventNet2021.06 | 75.2 | — | |
| H-First2021.06 | 71.2 | — | |
| BOHNASTraining protocol=NB201†2022.11 | — | 4.57 | |
| ConvSynSize of Hidden Layer=8002022.04 | — | 1.43 | |
| LSTM2022.04 | — | 2.95 | |
| MAP-SNNSize of Hidden Layer=800, MSP=true, ConvSyn=true2022.04 | — | 1.06 | |
| MLPSize of Hidden Layer=8002022.04 | — | 2.2 | |
| MSPSize of Hidden Layer=8002022.04 | — | 1.11 | |
| Phased-LSTM2022.04 | — | 2.62 | |
| Previous bestTraining protocol=NB201†2022.11 | — | 7.4 | |
| Spiking-CNN2022.04 | — | 4.28 | |
| Spiking-MLPSize of Hidden Layer=100002022.04 | — | 8.13 | |
| Spiking-MLPSize of Hidden Layer=8002022.04 | — | 1.26 | |
| Spiking-MLPSize of Hidden Layer=500-5002022.04 | — | 1.6 | |
| SSPSize of Hidden Layer=8002022.04 | — | 1.6 | |
| STBPSize of Hidden Layer=8002022.04 | — | 1.22 |