AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Vision Transformers for Dense Prediction

René Ranftl, Alexey Bochkovskiy, Vladlen Koltun

ICCV 2021 · Architectures and efficiency › Attention and transformers · approach: Efficient and robust vision transformers · 2,692 citations (OpenAlex)

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 changedWhereMeasuredEffect
use ViT-Hybrid backbone instead of ResNet50 (ablation)DPT · Depth reduced validation setmean relative deviation: 0.0935 -> 0.0783Improved
use ViT-Large instead of ViT-Base (ablation)DPT · Depth reduced validation setmean relative deviation: 0.0819 -> 0.0778Improved
use ViT-Hybrid instead of ViT-Base (ablation)DPT · Depth reduced validation setmean relative deviation: 0.0819 -> 0.0783Improved
use distilled DeIT-Base-Dist pretraining instead of ViT-Base (ablation)DPT · Depth reduced validation setmean relative deviation: 0.0819 -> 0.0796Improved
tap ResNet low-level stages R0,R1 instead of transformer layers (ablation)DPT-Hybrid · Depth reduced validation setmean relative deviation: 0.0787 -> 0.0733Improved
tap shallow and deep layers {3,6,9,12} instead of {9,10,11,12} (ablation)DPT-Base · Depth reduced validation setmean relative deviation: 0.0828 -> 0.0822Improved
project the readout token instead of ignoring it (ablation)DPT-Base · Depth reduced validation setmean relative deviation: 0.0822 -> 0.0819Improved
add the readout token to the tokens instead of ignoring it (ablation)DPT-Base · Depth reduced validation setmean relative deviation: 0.0822 -> 0.0831Hurt
train on larger MIX 6 dataset with the MiDaS convolutional networkMiDaS · Zero-shot depthDIW WHDR: 12.46 -> 12.95Improved
replace MiDaS conv network with DPT-Large on MIX 6MiDaS · Zero-shot depthDIW WHDR: 12.95 -> 10.82Improved
use DPT-Large instead of DPT-Hybrid for ADE20KDPT · ADE20K valmIoU: 49.02 -> 47.63Hurt
use DPT-Hybrid for monocular depth zero-shot on TUM instead of MiDaSMiDaS · TUM (zero-shot)delta>1.25 error: 12.51 -> 10.89Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy