AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Multiscale Vision Transformers

Haoqi Fan, Bo Xiong, Karttikeya Mangalam, Yanghao Li, Zhicheng Yan, Jitendra Malik, Christoph Feichtenhofer

ICCV 2021 · Architectures and efficiency › Attention and transformers · approach: Efficient and robust vision transformers · 1,594 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
add a single Q-pooling scale stage (resolution reduction) after 6 blocks (ablation)ViT-B · Kinetics-400 valtop-1 accuracy: 68.5 -> 71.0Improved
add K,V pooling (ablation)ViT-B · Kinetics-400 valtop-1 accuracy: 68.5 -> 69.1Improved
use learnable conv pooling instead of max pooling (ablation)MViT-B 16x4 · Kinetics-400 valtop-1 accuracy: 77.2 -> 78.3Improved
use average pooling instead of max pooling (ablation)MViT-B 16x4 · Kinetics-400 valtop-1 accuracy: 77.2 -> 75.4Hurt
use overlapping pooling kernel s+1 instead of kernel equal to stride (ablation)MViT-B 16x4 · Kinetics-400 valtop-1 accuracy: 76.1 -> 77.2Improved
add layer-normalized skip connection at stage transitions instead of unnormalized (ablation)MViT-B 16x4 · Kinetics-400 valtop-1 accuracy: 74.6 -> 77.2Improved
add skip connection at stage transitions (vs none) (ablation)MViT-B 16x4 · Kinetics-400 valtop-1 accuracy: 74.7 -> 77.2Improved
add separate space and time positional embedding instead of none (ablation)MViT-B 16x4 · Kinetics-400 valtop-1 accuracy: 75.8 -> 77.2Improved
use adaptive K,V pooling stride instead of fixed 2x4x4 (ablation)MViT-B 16x4 · Kinetics-400 valtop-1 accuracy: 74.8 -> 77.1Improved
add K,V pooling (adaptive 1x8x8) instead of none (ablation)MViT-B 16x4 · Kinetics-400 valtop-1 accuracy: 77.6 -> 77.2No clear effect
use temporal cubes (c_T=3) instead of 1x7x7 patches in cube1 (ablation)MViT-B · Kinetics-400 valtop-1 accuracy: 75.6 -> 75.9Improved
sample 16 frames with temporal stride 2 instead of 8 frames (ablation)MViT-B · Kinetics-400 valtop-1 accuracy: 75.9 -> 77.2Improved

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