Research map / Image generation / Super-resolution
EfficientViT: Lightweight Multi-Scale Attention for High-Resolution Dense Prediction
Han Cai, Junyan Li, Muyan Hu, Chuang Gan, Song Han
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add both multi-scale learning and global attention to the module (ablation) | EfficientViT module · Cityscapes (from scratch, equal MACs) | mIoU: 68.1 -> 74.5 | Improved |
| add one of the two components (multi-scale or global attention) alone (ablation) | EfficientViT module · Cityscapes (from scratch, equal MACs) | mIoU: 68.1 -> 72.3 and 68.1 -> 72.2 | Improved |
| replace SegFormer-B1 with EfficientViT-B1 | semantic segmentation model · Cityscapes val | mIoU: 78.5 -> 80.5 | Improved |
| replace SegNeXt-T with EfficientViT-B1 | semantic segmentation model · Cityscapes val | Orin latency: 93.2ms -> 24.3ms | Improved |
| replace SegFormer-B1 with EfficientViT-B1 | semantic segmentation model · ADE20K val | mIoU: 42.2 -> 42.8 | Improved |
| replace SegNeXt-S with EfficientViT-B2 | semantic segmentation model · ADE20K val | mIoU: 44.3 -> 45.9 | Improved |
| use EfficientViT-B3 instead of EfficientNet-B6 for classification | image classifier · ImageNet-1k val | top-1 accuracy: 84.0 -> 84.2 | Improved |
| replace Restormer with EfficientViT for super-resolution | super-resolution model · FFHQ super-resolution (512x512 to 1024x1024) | PSNR: 43.43 -> 43.58 | Improved |
| replace Restormer with EfficientViT for lightweight super-resolution | super-resolution model · BSD100 super-resolution (160x240 to 320x480) | PSNR: 32.31 -> 32.33 | No clear effect |
| replace softmax attention with ReLU-based linear attention (ablation) | attention module · mobile CPU latency (Snapdragon 855) | latency speedup: 3.3-4.5x faster | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Restormer — Restormer: Efficient Transformer for High-Resolution Image Restoration (CVPR 2022) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- EfficientViT: Memory Efficient Vision Transformer With Cascaded Group Attention (CVPR 2023 · similarity 0.90)
- AiluRus: A Scalable ViT Framework for Dense Prediction (NeurIPS 2023 · similarity 0.88)
- FasterViT: Fast Vision Transformers with Hierarchical Attention (ICLR 2024 · similarity 0.87)
- RepViT: Revisiting Mobile CNN From ViT Perspective (CVPR 2024 · similarity 0.86)
- SHViT: Single-Head Vision Transformer with Memory Efficient Macro Design (CVPR 2024 · similarity 0.85)
- ViT-CoMer: Vision Transformer with Convolutional Multi-scale Feature Interaction for Dense Predictions (CVPR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
