Efficient Vits On Imagenet 1K With Deit S

评估指标

GFLOPs

Top 1 Accuracy

评测结果

各个模型在此基准测试上的表现结果

			Paper Title	Repository
Base (DeiT-S)	4.6	79.8	Training data-efficient image transformers & distillation through attention
EViT (90%)	4.0	79.8	Not All Patches are What You Need: Expediting Vision Transformers via Token Reorganizations
DynamicViT (90%)	4.0	79.8	DynamicViT: Efficient Vision Transformers with Dynamic Token Sparsification
SPViT (3.9G)	3.9	79.8	SPViT: Enabling Faster Vision Transformers via Soft Token Pruning
LTMP (80%)	3.8	79.8	Learned Thresholds Token Merging and Pruning for Vision Transformers
A-ViT	3.6	78.6	AdaViT: Adaptive Tokens for Efficient Vision Transformer
EViT (80%)	3.5	79.8	Not All Patches are What You Need: Expediting Vision Transformers via Token Reorganizations
ToMe ($r=8$)	3.4	79.7	Token Merging: Your ViT But Faster
DynamicViT (80%)	3.4	79.8	DynamicViT: Efficient Vision Transformers with Dynamic Token Sparsification
SPViT	3.3	78.3	Pruning Self-attentions into Convolutional Layers in Single Path
IA-RED$^2$	3.2	79.1	IA-RED$^2$: Interpretability-Aware Redundancy Reduction for Vision Transformers	-
S$^2$ViTE	3.2	79.2	Chasing Sparsity in Vision Transformers: An End-to-End Exploration
BAT (70%)	3.0	79.6	Beyond Attentive Tokens: Incorporating Token Importance and Diversity for Efficient Vision Transformers
AS-DeiT-S (65%)	3.0	79.6	Adaptive Sparse ViT: Towards Learnable Adaptive Token Pruning by Fully Exploiting Self-Attention
EViT (70%)	3.0	79.5	Not All Patches are What You Need: Expediting Vision Transformers via Token Reorganizations
LTMP (60%)	3.0	79.6	Learned Thresholds Token Merging and Pruning for Vision Transformers
EvoViT	3.0	79.4	Evo-ViT: Slow-Fast Token Evolution for Dynamic Vision Transformer
eTPS	3.0	79.7	Joint Token Pruning and Squeezing Towards More Aggressive Compression of Vision Transformers
dTPS	3.0	80.1	Joint Token Pruning and Squeezing Towards More Aggressive Compression of Vision Transformers
DynamicViT (70%)	2.9	79.3	DynamicViT: Efficient Vision Transformers with Dynamic Token Sparsification

0 of 41 row(s) selected.