AI Paper Atlas

Research map / Video / Video segmentation and tracking

UniFormerV2: Unlocking the Potential of Image ViTs for Video Understanding

Kunchang Li, Yali Wang, Yinan He, Yizhuo Li, Yi Wang, Limin Wang, Yu Qiao

ICCV 2023 · Video › Video segmentation and tracking · approach: Self-supervised video representation learning · 94 citations (OpenAlex)

Read the paper

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 global UniBlock to CLIP-ViT baseline (ablation)UniFormerV2 · Kinetics-400top-1 accuracy: 83.1 -> 84.4Improved
add local UniBlock to CLIP-ViT baseline (ablation)UniFormerV2 · Something-Something V2top-1 accuracy: 45.1 -> 67.7Improved
add temporal downsampling with doubled frames (ablation)UniFormerV2 · Something-Something V2top-1 accuracy: 68.7 -> 69.5Improved
use local MHRA instead of ST-Adapter (ablation)local UniBlock · Something-Something V2top-1 accuracy: 68.0 -> 69.1Improved
use convolution instead of temporal self-attention for local temporal modeling (ablation)local UniBlock · Something-Something V2top-1 accuracy: 65.2 -> 67.5Improved
add DPE to local MHRA (ablation)local UniBlock · Something-Something V2top-1 accuracy: 69.1 -> 69.1No clear effect
use sequential fusion rather than parallel or hierarchical KV (ablation)multi-stage fusion · Something-Something V2top-1 accuracy: 69.1/68.9 -> 69.5Improved
use more queries (4 or 16) in global block (ablation)global UniBlock · Something-Something V2top-1 accuracy: 69.5 -> 69.1 (4 queries), 68.6 (16 queries)Hurt
post-pretrain on Kinetics-710 before finetuning (ablation)UniFormerV2 · Kinetics-400top-1 accuracy: 84.4 -> 85.6Improved
K710 training instead of K400+K600+K700 co-training (ablation)UniFormerV2 · Kinetics-700top-1 accuracy: -0.2 for co-training vs +0.5 for K710Improved
apply UniFormerV2 design to different pretrained ViTs (ablation)image ViTs · Kinetics-400 / SSV2top-1 accuracy: CLIP 84.4 / 69.5 vs TimeSformer 78.7 / 59.5Improved
UniFormerV2-L versus VideoMAE-L on SSV2UniFormerV2-L · Something-Something V2top-1 accuracy: 73.0 vs 74.3Hurt

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