AI Paper Atlas

Research map / Probabilistic and scientific modeling / Molecules and proteins

"SEA-RAFT: Simple, Efficient, Accurate RAFT for Optical Flow"

Yihan Wang, Lahav O Lipson, Jia Deng

ECCV 2024 · Probabilistic and scientific modeling › Molecules and proteins · approach: Flow matching generative modeling · 135 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
replace L1 loss with mixture-of-Laplace loss (beta1=0) (ablation)SEA-RAFT(S) · Spring subvalEPE: 0.206 -> 0.187Improved
use naive Laplace regression instead of L1 (ablation)SEA-RAFT(S) · Spring subvalEPE: 0.206 -> 0.217Hurt
do not fix beta1=0 in mixture of Laplace (ablation)SEA-RAFT(S) · Spring subvalEPE: 0.187 -> 0.248Hurt
use mixture of Gaussian instead of mixture of Laplace (ablation)SEA-RAFT(S) · Spring subvalEPE: 0.187 -> 0.210Hurt
add direct regression of initial flow (ablation)SEA-RAFT(S) · Spring subvalEPE: 0.201 -> 0.187Improved
add TartanAir rigid-flow pre-training (ablation)SEA-RAFT(S) · Spring subvalEPE: 0.187 -> 0.179Improved
add TartanAir pre-training (ablation)SEA-RAFT(M) · Spring train 1px1px outlier rate: 4.060 -> 3.888Improved
use ImageNet pre-trained backbones (ablation)SEA-RAFT(S) · Spring subvalEPE: 0.194 -> 0.187Improved
replace RAFT GRU with two ConvNeXt blocks (ablation)SEA-RAFT(S) · Spring subvalEPE and MACs: EPE 0.189 -> 0.187; 297.9G -> 284.7G MACsImproved
use four ConvNeXt blocks instead of two (ablation)SEA-RAFT(S) · Spring subvalEPE: 0.187 -> 0.189No clear effect
add TartanAir pre-training (ablation)SEA-RAFT(S) · Sintel train final (zero-shot)EPE: 4.32 -> 3.74Improved
add small amount of KITTI and HD1K data (ablation)SEA-RAFT(S) · Sintel train finalEPE: 4.32 -> 2.95Improved

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