AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

Learning by Distillation: A Self-Supervised Learning Framework for Optical Flow Estimation

Pengpeng Liu, Michael Rung-Tsong Lyu, Irwin King, Jia Xu

TPAMI 2021 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Pretext-task visual pre-training · 10 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 occlusion handling (ablation)DistillFlow · KITTI 2012 trainEPE-all: 7.33 -> 3.22Improved
add edge-aware smoothness (ablation)DistillFlow · KITTI 2012 trainEPE-all: 3.22 -> 2.92Improved
add data distillation (ablation)DistillFlow · KITTI 2012 trainEPE-all: 2.92 -> 1.46Improved
add data distillation (ablation)DistillFlow · KITTI 2015 trainEPE-all: 6.45 -> 3.20Improved
add model distillation (multi-teacher ensemble) (ablation)DistillFlow with data distillation · KITTI 2012 trainEPE-all: 1.46 -> 1.38Improved
add model distillation without data distillation (ablation)DistillFlow · KITTI 2012 trainEPE-all: 2.92 -> 2.86Improved
add data distillation (ablation)DistillFlow · Sintel Final rawEPE-all: 4.41 -> 3.98Improved
add knowledge distillation (ablation)FlowNetS · KITTI 2015 trainEPE-all: 8.85 -> 6.33Improved
add knowledge distillation (ablation)FlowNetC · KITTI 2015 trainEPE-all: 8.11 -> 5.47Improved
add semi-supervised distillation with self-annotated data (ablation)DistillFlow fine-tuning · KITTI 2015 validationEPE-all: 1.50 -> 1.44Improved
use distillation v4 instead of v1 (DDFlow) (ablation)DistillFlow · KITTI 2015 trainEPE-all: 3.88 -> 2.93Improved
use census transform instead of pixel brightness photometric loss (ablation)DistillFlow · KITTI 2012 trainEPE-all: 6.58 -> 3.22Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most 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

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