SPViT

Reported on 6 benchmarks across 1 task · 2 papers

Note: results are matched by exact model name. Different papers may use the same name for different model variants.

Computer Vision6 results

Image ClassificationonImageNet-1K (With LV-ViT-S)
GFLOPs· 2021-12-27
4.3
best: 6.6 (Base (LV-ViT-S))
SPViT: Enabling Faster Vision Transformers via Soft Token Pruning arXiv:2112.13890
Image ClassificationonImageNet-1K (With LV-ViT-S)
Top 1 Accuracy· 2021-12-27
83.1
best: 83.5 (MCTF ($r=8$))
SPViT: Enabling Faster Vision Transformers via Soft Token Pruning arXiv:2112.13890
Image ClassificationonImageNet-1K (with DeiT-S)
GFLOPs· 2021-11-23
3.3
best: 4.6 (Base (DeiT-S))
Pruning Self-attentions into Convolutional Layers in Single Path arXiv:2111.11802
Image ClassificationonImageNet-1K (with DeiT-S)
Top 1 Accuracy· 2021-11-23
78.3
best: 80.1 (MCTF ($r=16$))
Pruning Self-attentions into Convolutional Layers in Single Path arXiv:2111.11802
Image ClassificationonImageNet-1K (with DeiT-T)
GFLOPs· 2021-11-23
1
best: 1.2 (Base (DeiT-T))
Pruning Self-attentions into Convolutional Layers in Single Path arXiv:2111.11802
Image ClassificationonImageNet-1K (with DeiT-T)
Top 1 Accuracy· 2021-11-23
70.7
best: 72.9 (dTPS)
Pruning Self-attentions into Convolutional Layers in Single Path arXiv:2111.11802