AI Paper Atlas

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

Self-Distilled Feature Aggregation for Self-Supervised Monocular Depth Estimation

Zhengming Zhou, Qiulei Dong

ECCV 2022 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Pretext-task visual pre-training · 36 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
use modified Swin-transformer with high-resolution features (ablation)Swin+Raw · KITTI raw EigenAbs Rel: 0.102 -> 0.100Improved
replace first decoder block with SDFA (ablation)Swin-dagger+Raw · KITTI raw EigenAbs Rel: 0.100 -> 0.097Improved
replace three decoder blocks with SDFA (ablation)Swin-dagger+Raw · KITTI raw EigenAbs Rel: 0.100 -> 0.094Improved
use SDFA with self-distilled offsets instead of offset-based aggregation (ablation)Swin-dagger+OA · KITTI raw EigenRMSE: 4.082 -> 3.896Improved
use offset-based aggregation instead of concatenation (ablation)Swin-dagger+Raw · KITTI raw EigenAbs Rel: 0.100 -> 0.091Mixed
train OA with straightforward self-distillation (ablation)Swin-dagger+OA · KITTI raw EigenAbs Rel: 0.091 -> 0.092No clear effect
add photometric principle mask (ablation)SDFA-Net · KITTI raw EigenAbs Rel: 0.096 -> 0.090Improved
add visible principle mask (ablation)SDFA-Net · KITTI raw EigenAbs Rel: 0.090 -> 0.090No clear effect
add both principle masks (ablation)SDFA-Net · KITTI raw EigenAbs Rel: 0.095 -> 0.090Improved
add feature flipping in self-distilled propagation (ablation)SDFA-Net · KITTI raw EigenAbs Rel: 0.097 -> 0.090Improved
use SDFA with ResNet18 encoder instead of concatenation (ablation)Res18+Raw · KITTI raw EigenAbs Rel: 0.102 -> 0.101Improved
add Cityscapes to KITTI trainingSDFA-Net (PP) · KITTI raw EigenAbs Rel: 0.089 -> 0.084Improved

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