Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Image augmentation with invertible networks in interactive satellite image change detection

Hichem Sahbi

TL;DR

The paper tackles user-specific satellite image change detection under scarce labeled data by coupling a frugal, interactive Q&A loop with an invertible-network–based data augmentation strategy. It introduces a framework where an oracle labels a small display set and the change criterion is updated iteratively, while an invertible network maps data to a latent space to perform stable, near-linear augmentations that can be inverted back to the input space. The network is trained to be discriminative and stable, using an orthonormal-weight constraint to ensure invertibility, and augmentations are applied in latent space via unary and binary transformations. Experiments on the Jefferson dataset show that latent-space augmentation within the interactive loop yields superior performance, particularly in frugal labeling regimes, compared to several baselines.

Abstract

This paper devises a novel interactive satellite image change detection algorithm based on active learning. Our framework employs an iterative process that leverages a question-and-answer model. This model queries the oracle (user) about the labels of a small subset of images (dubbed as display), and based on the oracle's responses, change detection model is dynamically updated. The main contribution of our framework resides in a novel invertible network that allows augmenting displays, by mapping them from highly nonlinear input spaces to latent ones, where augmentation transformations become linear and more tractable. The resulting augmented data are afterwards mapped back to the input space, and used to retrain more effective change detection criteria in the subsequent iterations of active learning. Experimental results demonstrate superior performance of our proposed method compared to the related work.

Image augmentation with invertible networks in interactive satellite image change detection

TL;DR

The paper tackles user-specific satellite image change detection under scarce labeled data by coupling a frugal, interactive Q&A loop with an invertible-network–based data augmentation strategy. It introduces a framework where an oracle labels a small display set and the change criterion is updated iteratively, while an invertible network maps data to a latent space to perform stable, near-linear augmentations that can be inverted back to the input space. The network is trained to be discriminative and stable, using an orthonormal-weight constraint to ensure invertibility, and augmentations are applied in latent space via unary and binary transformations. Experiments on the Jefferson dataset show that latent-space augmentation within the interactive loop yields superior performance, particularly in frugal labeling regimes, compared to several baselines.

Abstract

This paper devises a novel interactive satellite image change detection algorithm based on active learning. Our framework employs an iterative process that leverages a question-and-answer model. This model queries the oracle (user) about the labels of a small subset of images (dubbed as display), and based on the oracle's responses, change detection model is dynamically updated. The main contribution of our framework resides in a novel invertible network that allows augmenting displays, by mapping them from highly nonlinear input spaces to latent ones, where augmentation transformations become linear and more tractable. The resulting augmented data are afterwards mapped back to the input space, and used to retrain more effective change detection criteria in the subsequent iterations of active learning. Experimental results demonstrate superior performance of our proposed method compared to the related work.
Paper Structure (9 sections, 3 equations, 1 figure, 2 tables)

This paper contains 9 sections, 3 equations, 1 figure, 2 tables.

Table of Contents

  1. Introduction
  2. Proposed Method
  3. Frugal satellite image change detection
  4. Invertible networks
  5. Augmentation
  6. Experiments
  7. Model Analysis & Ablation
  8. Comparison
  9. Conclusion

Figures (1)

  • Figure 1: This figure shows a comparison of different sample selection strategies w.r.t. different iterations (Iter) and the underlying sampling rates in table \ref{['tab12']} (Samp). Here Uncer and Rand stand for uncertainty and random sample selection respectively. Note that fully-supervised learning achieves an EER of $0.94 \%$. Related work stands for the method in refff33333; see again section \ref{['compare']} for more details. "Our (with augmentation)" stands for the best setting in table \ref{['tab12']} and "Our (without augmentation)" stands for the same setting but without augmentation.