AI Paper Atlas

Research map / Time series / Foundation models for time series

Panoptic Segmentation of Satellite Image Time Series With Convolutional Temporal Attention Networks

Vivien Sainte Fare Garnot, Loic Landrieu

ICCV 2021 · Time series › Foundation models for time series · approach: Satellite image time series · 283 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 interpolated multi-level temporal attention masks instead of temporal average of skip connections (ablation)U-TAE · PASTIS semantic segmentation (5-fold CV)mIoU: 63.1 vs 58.3Improved
add 1x1 conv after temporal average -> still worse than attention masks (ablation)U-TAE · PASTIS semantic segmentation (5-fold CV)mIoU: 63.1 vs 58.9Improved
add channel grouping of attention masks (G heads) instead of one averaged mask (ablation)U-TAE · PASTIS semantic segmentation (5-fold CV)mIoU: 63.1 vs 60.1Improved
use group normalization instead of batch normalization in encoder (ablation)U-TAE · PASTIS semantic segmentation (5-fold CV)mIoU: 63.1 vs 36.0Improved
use full time series instead of single August date (ablation)U-TAE · PASTIS semantic segmentation (5-fold CV)mIoU: 63.1 vs 28.3Improved
use full time series instead of single May date (ablation)U-TAE · PASTIS semantic segmentation (5-fold CV)mIoU: 63.1 vs 20.6Improved
replace U-BiConvLSTM-style recurrent encoder with U-TAE in PaPsPaPs · PASTIS panoptic segmentationPQ: 35.6 -> 43.8Improved
vary shape patch size S from 16 to 24 (ablation)PaPs · PASTIS panoptic segmentationPQ: 43.8 -> 41.3No clear effect
vary shape patch size S from 16 to 8 (ablation)PaPs · PASTIS panoptic segmentationPQ: 43.8 -> 42.7No clear effect
add residual CNN shape prediction instead of element-wise multiplication (ablation)PaPs · PASTIS panoptic segmentationSQ: 74.6 -> 81.5Improved
use time series instead of single image (ablation)PaPs · PASTIS panoptic segmentationPQ: 14.1 -> 43.8Improved
U-TAE vs strongest baseline 3D-Unetsemantic seg · PASTIS semantic segmentation (5-fold CV)mIoU: 58.4 -> 63.1Improved

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