Satellite-to-street view synthesis aims at generating a realistic street-view image from its corresponding satellite-view image. Although stable diffusion models have exhibited remarkable performance in a variety of image generation applications, their reliance on similar-view inputs to control the generated structure or texture restricts their application to the challenging cross-view synthesis task. In this work, we propose CrossViewDiff, a cross-view diffusion model for satellite-to-street view synthesis. To address the challenges posed by the large discrepancy across views, we design the satellite scene structure estimation and cross-view texture mapping modules to construct the structural and textural controls for street-view image synthesis. We further design a cross-view control guided denoising process that incorporates the above controls via an enhanced cross-view attention module. To achieve a more comprehensive evaluation of the synthesis results, we additionally design a GPT-based scoring method as a supplement to standard evaluation metrics. We also explore the effect of different data sources (e.g., text, maps, building heights, and multi-temporal satellite imagery) on this task. Results on three public cross-view datasets show that CrossViewDiff outperforms current state-of-the-art on both standard and GPT-based evaluation metrics, generating high-quality street-view panoramas with more realistic structures and textures across rural, suburban, and urban scenes. The code and models of this work will be released at https://opendatalab.github.io/CrossViewDiff/.
Fine-grained facial expression transfer from humans to humanoid agents presents a unique pattern recognition challenge due to the significant domain gap between biological facial dynamics and mechanical control spaces. While visual synthesis of talking heads has advanced rapidly, mapping high-dimensional visual cues to precise, physically constrained actuation signals remains an open problem, primarily due to the lack of large-scale paired data. To bridge this gap, we introduce X2C, a comprehensive benchmark dataset comprising 100,000 (image, control value) pairs. Unlike existing resources, X2C features nuanced, physically grounded expressions annotated with 30 continuous control parameters, establishing a high-fidelity standard for this task. Building on this resource, we propose X2CNet, a two-stage deep learning framework that explicitly decouples visual motion features from mechanical control regression to model the correspondence between human perceptual cues and humanoid actuation. Extensive experiments, including quantitative benchmarking and real-world physical validation, demonstrate that our approach achieves superior cross-domain consistency and enables robust, in-the-wild expression imitation. Video demonstrations are available at: https://lipzh5.github.io/X2CNet/.
Drug resistance remains a major challenge to successful cancer treatment. Although cell-autonomous genetic, molecular, and cellular modifications have historically been linked to resistance, there is growing evidence that extracellular vesicles (EVs) can promote the intercellular transfer of resistance-associated components. With a focus on the ways in which EV cargo modifies treatment response, this review critically investigates the role of EVs in therapeutic resistance in breast cancer, lung cancer, leukemia, glioblastoma, pancreatic cancer, and melanoma. The effects of major EV-associated factors on drug efflux, apoptosis, survival signaling, DNA repair, and metabolic adaptation are discussed. These factors include drug-efflux transporters, anti-apoptotic and oncogenic proteins, non-coding RNAs, DNA-repair molecules, metabolic components, and drug-sequestering molecules. As components of the tumor microenvironment that may strengthen treatment resistance, the review also considers EVs released by mesenchymal stromal cells, cancer-associated fibroblasts, and tumor-associated macrophages. By combining these results, we present a community-level view of EV-mediated resistance, where multidrug resistance and treatment failure may be caused by tumor-intrinsic resistance, horizontal transfer via tumor-derived EVs, and TME-derived EV signaling. Lastly, the potential of EVs as delivery systems, biomarkers, predictors, and therapeutic targets is examined, along with the present translational constraints and open issues. This synthesis emphasizes EV-mediated communication as a crucial addition to traditional cancer medication resistance models.
To construct the micro collision avoidance intelligent unit, efficient solutions have been pursued by modeling a specific collision-sensitive neuron, the lobula giant movement detector (LGMD), identified in the locust's visual neural system. However, the existing LGMD models cooperating with a threshold-based collision avoidance strategy are less robust in the presence of visual variability, including fluctuations in contrast and noise. Intriguingly, biological research underscores the potential of the single-peak modeling through the a function, suggesting a non-threshold collision avoidance strategy by identifying the peak response moment. Adapting the biological finding to the engineering applications, we establish a novel modeling method based on a-form perceptron, making the response curve behave like the a function. The single peak within the a-function-form response occurs before the collision, offering reliable collision warning cues. Numerical experiments substantiate the effectiveness of the proposed model against visual variability. This study extends the limits of the applications of LGMD models in diverse scenarios, and explores the potential of modeling neural computation in constructing micro-intelligent units.
