Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GCD-DDPM: A Generative Change Detection Model Based on Difference-Feature Guided DDPM

Yihan Wen, Xianping Ma, Xiaokang Zhang, Man-On Pun

TL;DR

This work proposes a generative CD model called GCD-DDPM to directly generate CD maps by exploiting the denoising diffusion probabilistic model (DDPM), instead of classifying each pixel into changed or unchanged categories.

Abstract

Deep learning (DL)-based methods have recently shown great promise in bitemporal change detection (CD). Existing discriminative methods based on Convolutional Neural Networks (CNNs) and Transformers rely on discriminative representation learning for change recognition while struggling with exploring local and long-range contextual dependencies. As a result, it is still challenging to obtain fine-grained and robust CD maps in diverse ground scenes. To cope with this challenge, this work proposes a generative change detection model called GCD-DDPM to directly generate CD maps by exploiting the Denoising Diffusion Probabilistic Model (DDPM), instead of classifying each pixel into changed or unchanged categories. Furthermore, the Difference Conditional Encoder (DCE), is designed to guide the generation of CD maps by exploiting multi-level difference features. Leveraging the variational inference (VI) procedure, GCD-DDPM can adaptively re-calibrate the CD results through an iterative inference process, while accurately distinguishing subtle and irregular changes in diverse scenes. Finally, a Noise Suppression-based Semantic Enhancer (NSSE) is specifically designed to mitigate noise in the current step's change-aware feature representations from the CD Encoder. This refinement, serving as an attention map, can guide subsequent iterations while enhancing CD accuracy. Extensive experiments on four high-resolution CD datasets confirm the superior performance of the proposed GCD-DDPM. The code for this work will be available at https://github.com/udrs/GCD.

GCD-DDPM: A Generative Change Detection Model Based on Difference-Feature Guided DDPM

TL;DR

This work proposes a generative CD model called GCD-DDPM to directly generate CD maps by exploiting the denoising diffusion probabilistic model (DDPM), instead of classifying each pixel into changed or unchanged categories.

Abstract

Deep learning (DL)-based methods have recently shown great promise in bitemporal change detection (CD). Existing discriminative methods based on Convolutional Neural Networks (CNNs) and Transformers rely on discriminative representation learning for change recognition while struggling with exploring local and long-range contextual dependencies. As a result, it is still challenging to obtain fine-grained and robust CD maps in diverse ground scenes. To cope with this challenge, this work proposes a generative change detection model called GCD-DDPM to directly generate CD maps by exploiting the Denoising Diffusion Probabilistic Model (DDPM), instead of classifying each pixel into changed or unchanged categories. Furthermore, the Difference Conditional Encoder (DCE), is designed to guide the generation of CD maps by exploiting multi-level difference features. Leveraging the variational inference (VI) procedure, GCD-DDPM can adaptively re-calibrate the CD results through an iterative inference process, while accurately distinguishing subtle and irregular changes in diverse scenes. Finally, a Noise Suppression-based Semantic Enhancer (NSSE) is specifically designed to mitigate noise in the current step's change-aware feature representations from the CD Encoder. This refinement, serving as an attention map, can guide subsequent iterations while enhancing CD accuracy. Extensive experiments on four high-resolution CD datasets confirm the superior performance of the proposed GCD-DDPM. The code for this work will be available at https://github.com/udrs/GCD.
Paper Structure (26 sections, 18 equations, 8 figures, 7 tables, 2 algorithms)

This paper contains 26 sections, 18 equations, 8 figures, 7 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related work
  3. CNN-based models
  4. Transformer-based models
  5. DDPM-based models
  6. METHODOLOGY
  7. Diffusion Process of GCD-DDPM
  8. Forward Process
  9. Reverse Process
  10. GCD-DDPM Network
  11. Noise Predictor
  12. DCE
  13. NSSE
  14. Experiment
  15. Experimental Dataset
  16. ...and 11 more sections

Figures (8)

  • Figure 1: A general comparison between (a) DDPM-CD and (b) the proposed GCD-DDPM. In DDPM-CD, DDPM is trained by remote sensing images in advance, and only the CD module is optimized in the CD task, while GCD-DDPM takes full advantage of the characteristics of the DDPM to train the entire model end-to-end. In the figure, the parameters of the blue blocks are fixed, while the orange blocks are optimized during the CD task.
  • Figure 2: An illustration of one timestep in the proposed GCD-DDPM, featuring Noise Predictor, NSSE modules and DCE. For clarity, only one NSSE module is presented in a gray box.
  • Figure 3: An illustration of the proposed attentive module NSSE. The spatial image information is first transformed into the frequency domain using the Fast Fourier Transform (FFT). After that, an attentive-like mechanism is utilized to suppress high-frequency noise before the extracted CD information is converted back into the spatial domain using the Inverse Fast Fourier Transform (IFFT).
  • Figure 4: Comparision of different state-of-the-art CD methods on CDD dataset: (a) Pre-change image, (b) Post-change image, (c) Ground-truth, (d) FC-SC, (e) SNUNet, (f) DT-SCN, (g) BIT, (h) AMTNet, (i) DDPM-CD, and (j) the proposed GCD-DDPM.
  • Figure 5: Comparision of different state-of-the-art CD methods on WHU-CD dataset: (a) Pre-change image, (b) Post-change image, (c) Ground-truth, (d) FC-SC, (e) SNUNet, (f) DT-SCN, (g) BIT, (h) AMTNet, (i) DDPM-CD, and (j) the proposed GCD-DDPM.
  • ...and 3 more figures