DynamicViT: Efficient Vision Transformers with Dynamic Token Sparsification
Attention is sparse in vision transformers. We observe the final prediction in vision transformers is only based on a subset of most informative tokens, which is sufficient for accurate image recognition. Based on this observation, we propose a dynamic token sparsification framework to prune redundant tokens progressively and dynamically based on the input. Specifically, we devise a lightweight prediction module to estimate the importance score of each token given the current features. The module is added to different layers to prune redundant tokens hierarchically. To optimize the prediction module in an end-to-end manner, we propose an attention masking strategy to differentiably prune a token by blocking its interactions with other tokens. Benefiting from the nature of self-attention, the unstructured sparse tokens are still hardware friendly, which makes our framework easy to achieve actual speed-up. By hierarchically pruning 66% of the input tokens, our method greatly reduces 31%~37% FLOPs and improves the throughput by over 40% while the drop of accuracy is within 0.5% for various vision transformers. Equipped with the dynamic token sparsification framework, DynamicViT models can achieve very competitive complexity/accuracy trade-offs compared to state-of-the-art CNNs and vision transformers on ImageNet. Code is available at https://github.com/raoyongming/DynamicViT
PDF Abstract NeurIPS 2021 PDF NeurIPS 2021 AbstractCode
Colab
Datasets
ImageNet
Results from the Paper
| Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Benchmark |
|---|---|---|---|---|---|---|
| Image Classification | ImageNet | DynamicViT-LV-M/0.8 | Top 1 Accuracy | 83.9 | # 347 | |
| Number of params | 57.1M | # 759 | ||||
| Hardware Burden | None | # 1 | ||||
| Operations per network pass | None | # 1 | ||||
| Efficient ViTs | ImageNet-1K (with DeiT-S) | DynamicViT (80%) | Top 1 Accuracy | 79.8 | # 4 | |
| GFLOPs | 3.4 | # 33 | ||||
| Efficient ViTs | ImageNet-1K (with DeiT-S) | DynamicViT (90%) | Top 1 Accuracy | 79.8 | # 4 | |
| GFLOPs | 4.0 | # 39 | ||||
| Efficient ViTs | ImageNet-1K (with DeiT-S) | DynamicViT (70%) | Top 1 Accuracy | 79.3 | # 25 | |
| GFLOPs | 2.9 | # 19 | ||||
| Efficient ViTs | ImageNet-1K (With LV-ViT-S) | DynamicViT (70%) | Top 1 Accuracy | 83.0 | # 10 | |
| GFLOPs | 4.6 | # 8 | ||||
| Efficient ViTs | ImageNet-1K (With LV-ViT-S) | DynamicViT (80%) | Top 1 Accuracy | 83.2 | # 5 | |
| GFLOPs | 5.1 | # 3 | ||||
| Efficient ViTs | ImageNet-1K (With LV-ViT-S) | DynamicViT (90%) | Top 1 Accuracy | 83.3 | # 3 | |
| GFLOPs | 5.8 | # 2 |
