Image Classification on ImageNet-1K (val) (Top-1 Accuracy and Delta)
71.71Delta Top-1 AccuracyWF
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| WFArchitecture=MobileNetV1, Sparsity Level=0.8, Dense Accuracy=71.952026.05 | 71.71 | 0.24 | |
| MPArchitecture=MobileNetV1, Sparsity Level=0.8, Dense Accuracy=71.952026.05 | 71.52 | 0.43 | |
| MPArchitecture=MobileNetV1, Sparsity Level=0.7, Dense Accuracy=71.952026.05 | 60.27 | 11.68 | |
| CBSArchitecture=MobileNetV1, Sparsity Level=0.8, Dense Accuracy=71.952026.05 | 55.57 | 16.38 | |
| FALCONArchitecture=MobileNetV1, Sparsity Level=0.8, Dense Accuracy=71.952026.05 | 46.13 | 25.82 | |
| WFArchitecture=MobileNetV1, Sparsity Level=0.7, Dense Accuracy=71.952026.05 | 42.59 | 29.36 | |
| CHITAArchitecture=MobileNetV1, Sparsity Level=0.8, Dense Accuracy=71.952026.05 | 42.17 | 29.78 | |
| BALF-P-0.7Backbone=MobileNetV2, ∆FLOPs%=−2.07±0.01, ∆Params%=−29.72±0.02, Evaluation Protocol=fine-tuning-free2025.09 | 28.08 | — | |
| MPArchitecture=MobileNetV1, Sparsity Level=0.6, Dense Accuracy=71.952026.05 | 24.8 | 47.15 | |
| L2+RESTBackbone=MobileNetV2, ∆Params%=−10.00, Evaluation Protocol=fine-tuning-free2025.09 | 18.47 | — | |
| LR-S2Backbone=MobileNetV2, ∆FLOPs%=−3.81, ∆Params%=−6.24, Evaluation Protocol=fine-tuning-free2025.09 | 17.46 | — | |
| ALDSBackbone=MobileNetV2, ∆FLOPs%=−2.62, ∆Params%=−37.61, Evaluation Protocol=fine-tuning-free2025.09 | 16.95 | — | |
| CBSArchitecture=MobileNetV1, Sparsity Level=0.7, Dense Accuracy=71.952026.05 | 16.84 | 55.11 | |
| SNOWSArchitecture=MobileNetV1, Sparsity Level=0.8, Dense Accuracy=70.892026.05 | 16.06 | 54.83 | |
| FALCONArchitecture=MobileNetV1, Sparsity Level=0.7, Dense Accuracy=71.952026.05 | 13.55 | 58.4 | |
| CHITAArchitecture=MobileNetV1, Sparsity Level=0.7, Dense Accuracy=71.952026.05 | 12.55 | 59.4 | |
| SVD-NASBackbone=MobileNetV2, ∆FLOPs%=−15.09, ∆Params%=−9.00, Evaluation Protocol=fine-tuning-free2025.09 | 12.54 | — | |
| STARFISHArchitecture=MobileNetV1, Sparsity Level=0.8, Dense Accuracy=71.952026.05 | 11.29 | 60.66 | |
| WFArchitecture=MobileNetV1, Sparsity Level=0.6, Dense Accuracy=71.952026.05 | 11.05 | 60.9 | |
| MPArchitecture=MobileNetV1, Sparsity Level=0.5, Dense Accuracy=71.952026.05 | 8.12 | 63.83 | |
| ScaleKDTeacher=BEIT-L/14‡, Student=Mixer-B/14, Teacher Params (M)=304.14, Student Params (M)=59.88, Teacher FLOPS (G)=81.06, Student FLOPS (G)=16.45, Pre-training=EVA [41]2024.11 | 6.27 | 82.89 | |
| BALF-P-0.75Backbone=MobileNetV2, ∆FLOPs%=−1.13±0.01, ∆Params%=−24.78±0.02, Evaluation Protocol=fine-tuning-free2025.09 | 6.17 | — | |
| SNOWSArchitecture=MobileNetV1, Sparsity Level=0.7, Dense Accuracy=70.892026.05 | 5.61 | 65.28 | |
| CBSArchitecture=MobileNetV1, Sparsity Level=0.6, Dense Accuracy=71.952026.05 | 5.58 | 66.37 | |
| ScaleKDTeacher=Swin-L†, Student=Mixer-B/16, Teacher Params (M)=196.53, Student Params (M)=59.88, Teacher FLOPS (G)=34.04, Student FLOPS (G)=12.61, Pre-training=IN-22K [45]2024.11 | 5.52 | 81.96 | |
| BALF-F-0.97Backbone=MobileNetV2, ∆FLOPs%=−2.51±0.00, ∆Params%=−1.55±0.00, Evaluation Protocol=fine-tuning-free2025.09 | 4.89 | — | |
| FALCONArchitecture=MobileNetV1, Sparsity Level=0.6, Dense Accuracy=71.952026.05 | 4.77 | 67.18 | |
| CHITAArchitecture=MobileNetV1, Sparsity Level=0.6, Dense Accuracy=71.952026.05 | 4.65 | 67.3 | |
| ScaleKDTeacher=Swin-L†, Student=Mixer-S/16, Teacher Params (M)=196.53, Student Params (M)=18.53, Teacher FLOPS (G)=34.04, Student FLOPS (G)=3.78, Pre-training=IN-22K [45]2024.11 | 4.61 | 78.63 | |
| L2+RESTBackbone=MobileNetV2, ∆Params%=−5.00, Evaluation Protocol=fine-tuning-free2025.09 | 4.42 | — | |
| ScaleKDTeacher=BEIT-L/14‡, Student=ViT-B/14, Teacher Params (M)=304.14, Student Params (M)=86.57, Teacher FLOPS (G)=81.06, Student FLOPS (G)=23.09, Pre-training=EVA [41]2024.11 | 4.41 | 86.43 | |
| STARFISHArchitecture=MobileNetV1, Sparsity Level=0.7, Dense Accuracy=71.952026.05 | 4.28 | 67.67 | |
| ScaleKDTeacher=Swin-L†, Student=ViT-S/16, Teacher Params (M)=196.53, Student Params (M)=22.05, Teacher FLOPS (G)=34.04, Student FLOPS (G)=4.61, Pre-training=IN-22K [45]2024.11 | 4.03 | 83.93 | |
| ScaleKDTeacher=Swin-L†, Student=ViT-B/16, Teacher Params (M)=196.53, Student Params (M)=86.57, Teacher FLOPS (G)=34.04, Student FLOPS (G)=17.58, Pre-training=IN-22K [45]2024.11 | 3.73 | 85.53 | |
| ScaleKDTeacher=BEIT-L/14‡, Student=ResNet-50, Teacher Params (M)=304.14, Student Params (M)=25.56, Teacher FLOPS (G)=81.06, Student FLOPS (G)=4.12, Pre-training=EVA [41]2024.11 | 3.7 | 82.34 | |
| ScaleKDTeacher=Swin-L†, Student=ResNet-50, Teacher Params (M)=196.53, Student Params (M)=25.56, Teacher FLOPS (G)=34.04, Student FLOPS (G)=4.12, Pre-training=IN-22K [45]2024.11 | 3.39 | 82.03 | |
| ScaleKDTeacher=Swin-L†, Student=MobileNet-V1, Teacher Params (M)=196.53, Student Params (M)=4.23, Teacher FLOPS (G)=34.04, Student FLOPS (G)=0.58, Pre-training=IN-22K [45]2024.11 | 3.05 | 75.15 | |
| WFArchitecture=MobileNetV1, Sparsity Level=0.5, Dense Accuracy=71.952026.05 | 3.04 | 68.91 | |
| ScaleKDTeacher=Swin-L†, Student=Swin-T, Teacher Params (M)=196.53, Student Params (M)=28.29, Teacher FLOPS (G)=34.04, Student FLOPS (G)=4.36, Pre-training=IN-22K [45]2024.11 | 2.62 | 83.8 | |
| SNOWSArchitecture=MobileNetV1, Sparsity Level=0.6, Dense Accuracy=70.892026.05 | 2.19 | 68.7 | |
| ScaleKDTeacher=Swin-L†, Student=ConvNeXt-T, Teacher Params (M)=196.53, Student Params (M)=28.59, Teacher FLOPS (G)=34.04, Student FLOPS (G)=4.46, Pre-training=IN-22K [45]2024.11 | 2.02 | 84.16 | |
| STARFISHArchitecture=MobileNetV1, Sparsity Level=0.6, Dense Accuracy=71.952026.05 | 1.85 | 70.1 | |
| CBSArchitecture=MobileNetV1, Sparsity Level=0.5, Dense Accuracy=71.952026.05 | 1.74 | 70.21 | |
| FALCONArchitecture=MobileNetV1, Sparsity Level=0.5, Dense Accuracy=71.952026.05 | 1.6 | 70.35 | |
| CHITAArchitecture=MobileNetV1, Sparsity Level=0.5, Dense Accuracy=71.952026.05 | 1.53 | 70.42 | |
| STARFISHArchitecture=MobileNetV1, Sparsity Level=0.5, Dense Accuracy=71.952026.05 | 0.83 | 71.12 | |
| SNOWSArchitecture=MobileNetV1, Sparsity Level=0.5, Dense Accuracy=70.892026.05 | 0.79 | 70.1 | |
| BALF-P-0.8Backbone=MobileNetV2, ∆FLOPs%=−0.23±0.00, ∆Params%=−19.80±0.00, Evaluation Protocol=fine-tuning-free2025.09 | 0.61 | — | |
| BALF-P-0.9Backbone=MobileNetV2, ∆FLOPs%=−0.12±0.00, ∆Params%=−10.69±0.05, Evaluation Protocol=fine-tuning-free2025.09 | 0.23 | — |