Remote sensing change detection (RSCD) aims to compare two or more images recorded for the same area but taken at different time stamps to quantitatively and qualitatively assess changes in geographical entities and environmental factors. Mainstream models usually built on pixel-by-pixel change detection paradigms, which cannot tolerate the diversity of changes due to complex scenes and variation in imaging conditions. Recently, some approaches have attempted to perform RSCD directly based on mask classification. However, they are inefficient and perform sub-optimally. In this paper, we provide an in-depth analysis of the change characteristic of RSCD, and further proposes the corresponding: 1) meta-architecture CDMask and 2) an enhanced instance network CDMaskFormer. Components of CDMask include Siamese backbone, change extractor, pixel decoder, transformer decoder and change customized normalized detector, which accommodates the change characteristics of the RSCD and ensures the proper functioning of the mask detection paradigm. Since the change query can be adaptively updated based on the bi-temporal feature content, the proposed CDMask can adapt to different latent data distributions, thus accurately identifying regions of interest changes in complex scenarios. Then, we further propose an enhanced instance network CDMaskFormer for the change detection task, which includes: (i) a Spatial-temporal convolutional attention-based instantiated change extractor to capture spatio-temporal context simultaneously with lightweight operations; and (ii) a scene-guided axial attention-instantiated transformer decoder to extract more spatial details. State-of-the-art performance of CDMaskFormer is achieved on five benchmark datasets with a satisfactory efficiency-accuracy trade-off. In particular, We construct a GAZACD dataset to explore the performance of CDMaskFormer in the real world. By analyzing 4353 pairs of images of the Gaza surface during and after the conflict, CDMaskFormer is able to effectively reveal the surface changes that occurred in the Gaza Strip and guide the post-disaster reconstruction. Code and dataset are available at https://github.com/xwmaxwma/rschange.

Generative Adversarial Networks (GANs) pose notable challenges in architectural design, as their performance is highly sensitive to hyperparameter tuning and network structure configurations. While Neural Architecture Search (NAS) offers promise for automated design, traditional NAS methods are hindered by exorbitant computational costs and insufficient diversity when navigating high-dimensional search spaces. Drawing on the equivalence between diffusion models and evolutionary processes, this work introduces a Consistency Diffusion Evolution algorithm tailored for GAN architecture search (CoDE-GAN). Our approach leverages probability flow matching theory to compress multi-step evolutionary sampling into a single step, drastically reducing search overhead while preserving solution diversity—a critical factor for discovering robust architectures. By integrating Frechet Inception Distance (FID) and Inception Score (IS) into the diffusion loss function Via Kernel Density Estimation (KDE), we achieve dynamic balance between generation quality and diversity throughout optimization. Extensive experiments on benchmark datasets validate that our method delivers competitive sample fidelity and architectural efficiency: on CIFAR-10, it attains the second-best IS (9.05) and a competitive FID (8.96); on CIFAR-100, it establishes new benchmarks with optimal IS (9.19) and FID (13.37); on STL-10, it achieves the highest IS (10.54) with a competitive FID (21.04). Meanwhile, it requires only 0.36 GPU days.