AI Paper Atlas

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

SatMAE: Pre-training Transformers for Temporal and Multi-Spectral Satellite Imagery

Yezhen Cong, Samar Khanna, Chenlin Meng, Patrick Liu, Erik Rozi, Yutong He, Marshall Burke, David B. Lobell, Stefano Ermon

NeurIPS 2022 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Pretext-task visual pre-training

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 temporal encoding (ablation)temporal SatMAE · fMoW RGB temporalTop-1 accuracy: 78.07 -> 79.69Improved
Use independent masking instead of consistent masking (ablation)temporal SatMAE · fMoW RGB temporalTop-1 accuracy: 78.45 -> 79.69Improved
Add consistent cropping (ablation)temporal SatMAE · fMoW RGB temporalTop-1 accuracy: 79.90 -> 79.69Hurt
Add test-time augmentation (ablation)temporal SatMAE · fMoW RGB temporalTop-1 accuracy: 79.69 -> 81.49Improved
Stack temporal sequence along channels instead of temporal SatMAE (ablation)SatMAE · fMoW RGB temporalTop-1 accuracy: 81.49 -> 75.85Hurt
Group bands by wavelength/resolution (X) instead of random split (R) (ablation)spectral SatMAE · fMoW SentinelTop-1 accuracy: 58.76 -> 59.30Improved
Group bands by wavelength (X) instead of halves (H) (ablation)spectral SatMAE · fMoW SentinelTop-1 accuracy: 57.78 -> 59.30Improved
Use independent masking instead of consistent masking (ablation)spectral SatMAE · fMoW SentinelTop-1 accuracy: 57.76 -> 59.30Improved
Add spectral group encoding (ablation)spectral SatMAE · fMoW SentinelTop-1 accuracy: 58.87 -> 59.30Improved
Use ViT-Large instead of ViT-Base (ablation)spectral SatMAE · fMoW SentinelTop-1 accuracy: 59.11 -> 59.30Improved
Lower masking ratio from 0.75 to 0.6 (ablation)temporal SatMAE · fMoW RGB temporalTop-1 accuracy: 7.19%Hurt
Pre-train SatMAE+Group+IM vs scratch ResNet-152SatMAE+Group+IM · fMoW SentinelTop-1 accuracy: 49.12 -> 59.30Improved

Later papers that compared with this method

4 later accepted papers report SatMAE as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.

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