AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

MaskFlownet: Asymmetric Feature Matching With Learnable Occlusion Mask

Shengyu Zhao, Yilun Sheng, Yue Dong, Eric I-Chao Chang, Yan Xu

CVPR 2020 · Label-efficient and robust learning › Recognition and long-tail · approach: Correspondence and causal representation learning · 200 citations (OpenAlex)

Check 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 standard FMM with OFMM using a learnable mask and trade-off term (ablation)MaskFlownet-S · Sintel clean and final validation, Chairs-trainedAEPE: 3.02 -> 3.06 (clean); 4.70 -> 4.52 (final)Mixed
use asymmetric deformable feature matching instead of symmetric FMM (ablation)MaskFlownet-S · Sintel clean and final validation, Chairs-trainedAEPE: 3.02 -> 2.70 (clean); 4.70 -> 4.07 (final)Improved
add symmetric convolution to OFMM (ablation)OFMM · Sintel clean and final train, Chairs-trainedAEPE: 3.20 -> 3.33 (clean); 4.50 -> 4.64 (final)Hurt
add asymmetric convolution before warping (ablation)OFMM · Sintel clean and final train, Chairs-trainedAEPE: 3.20 -> 2.96 (clean); 4.50 -> 4.29 (final)Improved
replace warping and extra convolution with deformable convolution (ablation)OFMM · Sintel clean and final train, Chairs-trainedAEPE: 3.20 -> 2.88 (clean); 4.50 -> 4.25 (final)Improved
add both the learnable mask and trade-off term (ablation)AsymOFMM · Sintel clean train, Chairs-trainedAEPE: 3.08 -> 2.88Improved
add the learnable mask without the trade-off term (ablation)AsymOFMM · Sintel clean and final train, Chairs-trainedAEPE: 3.08 -> 3.06 (clean); 4.29 -> 4.32 (final)Hurt
add the trade-off term without the learnable mask (ablation)AsymOFMM · Sintel clean and final train, Chairs-trainedAEPE: 3.08 -> 2.97 (clean); 4.29 -> 4.30 (final)Mixed
cascade using a single new feature pyramid without mask (ablation)MaskFlownet-S · Sintel val clean and finalAEPE: 2.70 -> 2.53 (clean); 4.07 -> 3.90 (final)Improved
use dual feature pyramids instead of a single pyramid without mask (ablation)cascaded MaskFlownet · Sintel val clean and finalAEPE: 2.53 -> 2.52 (clean); 3.90 -> 3.85 (final)Improved
concatenate the occlusion mask in the dual-pyramid cascade (ablation)cascaded MaskFlownet · Sintel val clean and finalAEPE: 2.52 -> 2.52 (clean); 3.85 -> 3.83 (final)Mixed
use MaskFlownet instead of VCNoptical flow estimation · MPI Sintel clean testAEPE: 2.81 -> 2.52Improved

Later papers that compared with this method

1 later accepted paper reports MaskFlowNet as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.

Earlier methods this paper compares with

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