AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

SatlasPretrain: A Large-Scale Dataset for Remote Sensing Image Understanding

Favyen Bastani, Piper Wolters, Ritwik Gupta, Joe Ferdinando, Aniruddha Kembhavi

ICCV 2023 · Vision-language and multimodal › Human and embodied modeling · approach: Vision-language pretraining · 231 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
fine-tune jointly trained SatlasNet on each label type instead of per-type training (ablation)SatlasNet · SatlasPretrain test, all label types and image modesaverage performance: 7.1% relative improvementImproved
use multi-image time series input instead of single image (ablation)SatlasNet · SatlasPretrain test, all label types and image modesaverage performance: 5.6% relative improvementImproved
joint training only without per-type fine-tuning (ablation)SatlasNet (single-image) · SatlasPretrain test, low-res Sentinel-2, segmentationF1: 64.8 -> 55.8Hurt
multi-image input without joint training (ablation)SatlasNet, high-res NAIP · SatlasPretrain test, high-res NAIP, propertiesaccuracy: 86.6 -> 77.3Hurt
pre-train on SatlasPretrain instead of ImageNetSwin-Base downstream fine-tuning · Seven downstream tasks, 50 examples (average)average accuracy/F1: 0.38 -> 0.56Improved
pre-train on SatlasPretrain instead of ImageNetSwin-Base downstream fine-tuning · Seven downstream tasks, full dataset (average)average accuracy/F1: 0.75 -> 0.83Improved
pre-train on SatlasPretrain instead of DOTASwin-Base, Airbus Ships · Airbus Ships, 50 examplesF1: 0.62 -> 0.56Hurt
use MoCo self-supervised pre-training on SatlasPretrain images instead of supervised pre-trainingSwin-Base downstream fine-tuning · Seven downstream tasks, 50 examples (average)average accuracy/F1: 0.56 -> 0.21Hurt
use remote-sensing-specific self-supervised SeCo instead of ImageNet pre-trainingSwin-Base downstream fine-tuning · Seven downstream tasks, 50 examples (average)average accuracy/F1: 0.38 -> 0.38No clear effect

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