# ChangeT3: Evaluating Vision Test-Time Training for Remote Sensing Change Detection

arXiv: https://arxiv.org/abs/2610.15813 · HTML: https://mlsanity.com/paper/2610.15813/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 20 of its 24 references that are in the map.
- First posted on arXiv 11 Oct 2026.

## Builds on (most cited first, 20 of 20)

- [Linear Transformers Are Secretly Fast Weight Programmers](https://mlsanity.com/paper/2102.11174/) (2021)
- [A Comprehensive Survey on Test-Time Adaptation under Distribution Shifts](https://mlsanity.com/paper/2303.15361/) (2023)
- [Learning to (Learn at Test Time): RNNs with Expressive Hidden States](https://mlsanity.com/paper/2407.04620/) (2024)
- [Titans: Learning to Memorize at Test Time](https://mlsanity.com/paper/2501.00663/) (2024)
- [Fully Convolutional Siamese Networks for Change Detection](https://mlsanity.com/paper/1810.08462/) (2018)
- [Remote Sensing Image Change Detection with Transformers](https://mlsanity.com/paper/2103.00208/) (2021)
- [A Transformer-Based Siamese Network for Change Detection](https://mlsanity.com/paper/2201.01293/) (2022)
- [Test-Time Training with Masked Autoencoders](https://mlsanity.com/paper/2209.07522/) (2022)
- [Test-Time Training Done Right](https://mlsanity.com/paper/2505.23884/) (2025)
- [ChangeMamba: Remote Sensing Change Detection With Spatiotemporal State Space Model](https://mlsanity.com/paper/2404.03425/) (2024)
- [Building Change Detection for Remote Sensing Images Using a Dual Task Constrained Deep Siamese Convolutional Network Model](https://mlsanity.com/paper/1909.07726/) (2019)
- [ViT3: Unlocking Test-Time Training in Vision](https://mlsanity.com/paper/2512.01643/) (2025)
- [Enhanced SCanNet with CBAM and Dice Loss for Semantic Change Detection](https://mlsanity.com/paper/2505.04199/) (2025)
- [Precision Spatio-Temporal Feature Fusion for Robust Remote Sensing Change Detection](https://mlsanity.com/paper/2507.11523/) (2025)
- [Beyond Quadratic: Linear-Time Change Detection with RWKV](https://mlsanity.com/paper/2603.19606/) (2026)
- [A Controlled Benchmark of Visual State-Space Backbones with Domain-Shift and Boundary Analysis for Remote-Sensing Segmentation](https://mlsanity.com/paper/2604.18721/) (2026)
- [Lightweight Structure-aware Transformer Network for VHR Remote Sensing Image Change Detection](https://mlsanity.com/paper/2306.01988/) (2023)
- [Mamba-FCS: Joint Spatio- Frequency Feature Fusion, Change-Guided Attention, and SeK Loss for Enhanced Semantic Change Detection in Remote Sensing](https://mlsanity.com/paper/2508.08232/) (2025)
- [Towards Remote Sensing Change Detection with Neural Memory](https://mlsanity.com/paper/2602.10491/) (2026)
- [MambaRefine-CD: MambaVision with Region-Boundary Temporal Refinement](https://mlsanity.com/paper/2607.04403/) (2026)

## Abstract (arXiv)

Bitemporal remote sensing change detection requires representations that preserve spatial detail while capturing contextual information relevant to changes between acquisitions. Visual test-time training offers an alternative to explicit token-to-token attention by adapting a compact inner model to each image. This letter describes ChangeT3, a vanilla Siamese configuration combining the existing hierarchical ViT3-Tiny encoder with the multiscale ChangeFormer decoder. The two image streams share learned encoder parameters, while their fast weights are computed independently from their respective spatial features. Each encoder block retains the inherited global simplified SwiGLU and local adaptive depthwise-convolution branches. Temporal comparison is performed by the decoder through learned fusion of paired features at four resolutions, followed by coarse-to-fine aggregation and binary prediction. This separation provides a direct setting for examining image-conditioned inner updates without introducing an additional temporal interaction module. Trained from scratch on LEVIR-CD+ with the same decoder, recipe, and budget as a ChangeFormer control, ChangeT3 reaches an F1 of 80.61±0.90 against 78.84±0.21 for ChangeFormer (three seeds each); as in the standard protocol for this dataset, checkpoints are selected on the test split. Trained with both inner updates disabled, the same encoder reaches 78.63±0.20. Neither model transfers zero-shot to SYSU-CD, DSIFN-CD, or WHU-CD, and ChangeT3 is the slower of the two at small inputs despite similar parameter and operation counts.