Image Classification on Tiny-ImageNet
88.42AccuracyFractalDB
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| FractalDBBackbone=ViT-B, Pre-training=ImageNet-1K, Initialization/Warm-up=FractalDB [23] warm-up, Evaluation Protocol=Fine-tuned2025.11 | 88.42 | — | — | |
| Procedural warm-upBackbone=ViT-B, Pre-training=ImageNet-1K, Initialization/Warm-up=Procedural warm-up (ours), Evaluation Protocol=Fine-tuned2025.11 | 87.93 | — | — | |
| Mimetic init.Backbone=ViT-B, Pre-training=ImageNet-1K, Initialization/Warm-up=Mimetic init. [51], Evaluation Protocol=Fine-tuned2025.11 | 87.29 | — | — | |
| Random init.Backbone=ViT-B, Pre-training=ImageNet-1K, Initialization/Warm-up=Random init. [10], Evaluation Protocol=Fine-tuned2025.11 | 86.59 | — | — | |
| SAFT-LEvaluation Protocol=Clean, Backbone=CLIP-L/142026.02 | 80.39 | — | — | |
| FAREEvaluation Protocol=Clean, Backbone=CLIP-L/142026.02 | 68.23 | — | — | |
| SAFT-LEvaluation Protocol=Robust, Backbone=CLIP-L/14, Adversarial Attack=PGD-100, Perturbation Budget (epsilon)=1/2552026.02 | 68.06 | — | — | |
| GradNDTarget Architecture=Swin-Tiny2026.03 | 67.19 | — | — | |
| SCOPETarget Architecture=Swin-Tiny2026.03 | 67.17 | — | — | |
| FedCSTarget Architecture=Swin-Tiny2026.03 | 66.85 | — | — | |
| EL2NTarget Architecture=Swin-Tiny2026.03 | 66.76 | — | — | |
| FedCoreTarget Architecture=Swin-Tiny2026.03 | 66.47 | — | — | |
| DeepReDuce (Baseline)Network-Dataset=DeepReD1-Tiny, #ReLUs (K)=917.52021.06 | 64.66 | 12.27 | — | |
| Circa (NegPass)Network-Dataset=DeepReD1-Tiny, #ReLUs (K)=917.5, NegPass bits=142021.06 | 64.62 | 6.68 | 1.8 | |
| Circa (PosZero)Network-Dataset=DeepReD1-Tiny, #ReLUs (K)=917.5, PosZero bits=142021.06 | 64.53 | 6.68 | 1.8 | |
| CV-DDBackbone=ResNet50, IPC (Images Per Class)=50, Ratio=10.0%2025.01 | 64.1 | — | — | |
| CutMixModel=ResNet2020.02 | 64.08 | — | — | |
| AdaBNTraining Augmentations=true2021.03 | 64 | — | — | |
| OursBackbone=ResNet-502026.03 | 63.7 | — | 1.26 | |
| Full-dataBackbone=ResNet-502026.03 | 63.5 | — | 1 | |
| InfoBatchBackbone=ResNet-502026.03 | 63.4 | — | 1.71 | |
| CV-DDBackbone=ResNet101, IPC (Images Per Class)=50, Ratio=10.0%2025.01 | 63.1 | — | — | |
| HexFormer-HybridArchitecture=ViT-Tiny, Pre-training=None2026.01 | 62.93 | — | — | |
| DeepReDuce (Baseline)Network-Dataset=DeepReD2-Tiny, #ReLUs (K)=458.82021.06 | 62.26 | 6.5 | — | |
| Circa (PosZero)Network-Dataset=DeepReD5-Tiny, #ReLUs (K)=393.2, PosZero bits=152021.06 | 61.66 | 3.21 | 1.7 | |
| DeepReDuce (Baseline)Network-Dataset=DeepReD5-Tiny, #ReLUs (K)=393.22021.06 | 61.65 | 5.38 | — | |
| Circa (NegPass)Network-Dataset=DeepReD5-Tiny, #ReLUs (K)=393.2, NegPass bits=152021.06 | 61.63 | 3.21 | 1.7 | |
| BaselineNetwork=ResNet18, #ReLUs (K)=2228.22021.06 | 61.6 | 44.55 | — | |
| FMixModel=ResNet2020.02 | 61.43 | — | — | |
| Circa (NegPass)Network-Dataset=DeepReD2-Tiny, #ReLUs (K)=458.8, NegPass bits=152021.06 | 61.28 | 3.94 | 1.6 | |
| Circa (PosZero)Network-Dataset=DeepReD2-Tiny, #ReLUs (K)=458.8, PosZero bits=152021.06 | 61.26 | 3.94 | 1.6 | |
| NRR-DDBackbone=ResNet-18, IPC=502025.03 | 61.2 | — | — | |
| CV-DDBackbone=ResNet18, IPC (Images Per Class)=50, Ratio=10.0%2025.01 | 61.2 | — | — | |
| JEPAMatchIteration=2^182026.04 | 61.18 | — | — | |
| HexFormerArchitecture=ViT-Tiny, Pre-training=None2026.01 | 60.87 | — | — | |
| Euclidean ViTArchitecture=ViT-Tiny, Pre-training=None2026.01 | 60.74 | — | — | |
| Circa (PosZero)Network=ResNet18, #ReLUs (K)=2228.2, Stochastic ReLU Mode=PosZero, Truncation Bits=122021.06 | 60.65 | 14.28 | 3.1 | |
| Circa (NegPass)Network=ResNet18, #ReLUs (K)=2228.2, Stochastic ReLU Mode=NegPass, Truncation Bits=132021.06 | 60.6 | 14.28 | 3.1 | |
| AdaBNTraining Augmentations=false2021.03 | 60.3 | — | — | |
| SoftMatchIteration=2^182026.04 | 59.91 | — | — | |
| SEALGeneration=Gen=102023.04 | 59.22 | — | — | |
| SEALEvaluation Protocol=linear probe, Generations (G)=102023.04 | 59.22 | — | — | |
| DeepReDuce (Baseline)Network-Dataset=DeepReD6-Tiny, #ReLUs (K)=229.42021.06 | 59.18 | 3.18 | — | |
| OursBackbone=VGG-162026.03 | 58.89 | — | 1.34 | |
| Circa (NegPass)Network-Dataset=DeepReD6-Tiny, #ReLUs (K)=229.4, NegPass bits=152021.06 | 58.65 | 2.01 | 1.6 | |
| Circa (PosZero)Network-Dataset=DeepReD6-Tiny, #ReLUs (K)=229.4, PosZero bits=152021.06 | 58.61 | 2.01 | 1.6 | |
| Full-precisionBackbone=ResNet-18, Source=Reported in [3]2021.12 | 58.35 | — | — | |
| FlexMatchIteration=2^182026.04 | 58.27 | — | — | |
| SEALGeneration=Gen=32023.04 | 58.25 | — | — | |
| Full-dataBackbone=VGG-162026.03 | 58.21 | — | 1 | |
| RDEDBackbone=ResNet-18, IPC=502025.03 | 58.2 | — | — | |
| RDEDBackbone=ResNet18, IPC (Images Per Class)=50, Ratio=10.0%2025.01 | 58.2 | — | — | |
| Procedural warm-up2025.11 | 58.2 | — | — | |
| Mimetic initialization2025.11 | 57.2 | — | — | |
| LLFGeneration=Gen=102023.04 | 56.92 | — | — | |
| LLFEvaluation Protocol=linear probe, Generations (G)=102023.04 | 56.92 | — | — | |
| CLIPBackbone=VGG-162026.03 | 56.34 | — | 1.1 | |
| LLFGeneration=Gen=32023.04 | 56.12 | — | — | |
| AdaSTEBackbone=ResNet-18, Mixup=true, Annealing=true2021.12 | 56.11 | — | — | |
| MixUpModel=ResNet2020.02 | 55.96 | — | — | |
| BaselineModel=ResNet2020.02 | 55.94 | — | — | |
| BayesBiNNBackbone=ResNet-18, Mixup=true2021.12 | 55.84 | — | — | |
| SCOPETarget Architecture=ResNet-502026.03 | 55.68 | — | — | |
| BaselineNetwork=ResNet32, #ReLUs (K)=1212.42021.06 | 55.53 | 24.24 | — | |
| Default random initialization2025.11 | 55.42 | — | — | |
| LPLDIPC=100, Pruning Ratio=1x, Backbone=ResNet-1012024.10 | 55.4 | — | — | |
| LPLDIPC=100, Pruning Ratio=1x, Backbone=ResNet-502024.10 | 55.3 | — | — | |
| FractalDB warm-up2025.11 | 55.17 | — | — | |
| Circa (NegPass)Network=ResNet32, #ReLUs (K)=1212.4, Stochastic ReLU Mode=NegPass, Truncation Bits=162021.06 | 55.15 | 9.04 | 2.7 | |
| FedCoreTarget Architecture=ResNet-502026.03 | 55.06 | — | — | |
| AdaSTEBackbone=ResNet-18, Mixup=false, Annealing=true2021.12 | 54.92 | — | — | |
| AdaSTEBackbone=ResNet-18, Annealing=true, Trials=52021.12 | 54.92 | — | — | |
| FedCSTarget Architecture=ResNet-502026.03 | 54.9 | — | — | |
| MD-tanhBackbone=ResNet-18, Source=Reported in [3]2021.12 | 54.62 | — | — | |
| Circa (PosZero)Network=ResNet32, #ReLUs (K)=1212.4, Stochastic ReLU Mode=PosZero, Truncation Bits=152021.06 | 54.56 | 9.04 | 2.7 | |
| NormalGeneration=Gen=12023.04 | 54.37 | — | — | |
| NormalEvaluation Protocol=linear probe, Generations (G)=12023.04 | 54.37 | — | — | |
| BayesBiNNBackbone=ResNet-18, Mixup=false2021.12 | 54.22 | — | — | |
| BayesBiNNBackbone=ResNet-18, Trials=52021.12 | 54.22 | — | — | |
| LPLDIPC=100, Pruning Ratio=10x, Backbone=ResNet-1012024.10 | 54.1 | — | — | |
| LPLDIPC=100, Pruning Ratio=10x, Backbone=ResNet-502024.10 | 54 | — | — | |
| CV-DDBackbone=ResNet101, IPC (Images Per Class)=10, Ratio=2.0%2025.01 | 53.9 | — | — | |
| LPLDIPC=100, Pruning Ratio=20x, Backbone=ResNet-1012024.10 | 53.7 | — | — | |
| SRe2L++Backbone=ResNet50, IPC (Images Per Class)=50, Ratio=10.0%2025.01 | 53.7 | — | — | |
| SRe2L++Backbone=ResNet101, IPC (Images Per Class)=50, Ratio=10.0%2025.01 | 53.7 | — | — | |
| CleanEvaluation Protocol=SL2024.02 | 53.5 | — | — | |
| SRe2L++Backbone=ResNet18, IPC (Images Per Class)=50, Ratio=10.0%2025.01 | 53.5 | — | — | |
| AdaSTEBackbone=ResNet-18, Annealing=false, Trials=52021.12 | 53.49 | — | — | |
| GradNDTarget Architecture=ResNet-502026.03 | 53.48 | — | — | |
| EL2NTarget Architecture=ResNet-502026.03 | 53.35 | — | — | |
| LViTArchitecture=ViT-Base, Pre-training=None2026.01 | 53.01 | — | — | |
| CV-DDBackbone=ResNet18, IPC (Images Per Class)=10, Ratio=2.0%2025.01 | 53 | — | — | |
| CV-DDBackbone=ResNet50, IPC (Images Per Class)=10, Ratio=2.0%2025.01 | 52.8 | — | — | |
| LPLDIPC=100, Pruning Ratio=20x, Backbone=ResNet-502024.10 | 52.7 | — | — | |
| MD-tanh-sBackbone=ResNet-18, Source=Reported in [3]2021.12 | 52.32 | — | — | |
| WaveMix-128/7#Param (Million)=2.42, GPU (GB)=1.32022.03 | 52.03 | — | — | |
| MD-softmax-sBackbone=ResNet-18, Source=Reported in [3]2021.12 | 51.81 | — | — | |
| PMFBackbone=ResNet-18, Source=Reported in [3]2021.12 | 51.52 | — | — | |
| WaveMix-256/7#Param (Million)=9.62, GPU (GB)=2.32022.03 | 51.37 | — | — | |
| NormalGeneration=Gen=32023.04 | 51.16 | — | — |