Recent advances in deep learning and computer vision have been driven by the creation and use of numerous new benchmark datasets and challenges in various fields of application. However, when deployed in real-world environments, these applications do not always demonstrate good performance. This is attributable to several factors. Benchmark datasets may not adequately represent all of the characteristics and limitations of a certain application, so that algorithms and models tuned on these datasets do not work well in out-of-distribution (OOD) tests, namely when the characteristics of the test data and environment change. Furthermore, practical engineering problems arise when deploying lab-trained models in real scenarios, such as scalability and computation on memory- and resource-limited edge devices, model compression or mixed floating-point calculations, and optimization for embedded devices, etc.
The main goal of this Research Topic is to present new contributions on recent advances in real-world AI and Computer Vision applications. This includes advances in the theory of AI model generalization, the deployment of models, and comparisons between lab-benchmarks and real-world performance. Authors are encouraged to highlight the pros and cons of new methods or established techniques in a specific scenario, focusing on theoretical advances in generalization, and both software and hardware configurations managing trade-offs between performance and feasibility. Contributions may encompass study cases, comparisons of techniques, validation protocols, and practical applications of computer vision and AI.
Research articles are welcome that highlight new approaches in an end-to-end manner, as are review articles surveying state-of-the-art methods in AI generalization and deployment in real-world scenarios. Topics of interest include, but are not limited to:
- Theoretical advances: distribution shifts, domain generalization, generalization, out-of-distribution detection, shortcut learning transfer learning
- Computer Vision and AI tasks: anomaly detection, embedded and edge computing, image and video generation, object detection, classification and tracking, semantic segmentation, remote sensing, scene understanding, visual saliency
- Application scenarios: autonomous vehicles, bio-medical image analysis, art and cultural heritage, healthcare and medicine, marketing and retail, (social) robotics, intelligent surveillance
Recent advances in deep learning and computer vision have been driven by the creation and use of numerous new benchmark datasets and challenges in various fields of application. However, when deployed in real-world environments, these applications do not always demonstrate good performance. This is attributable to several factors. Benchmark datasets may not adequately represent all of the characteristics and limitations of a certain application, so that algorithms and models tuned on these datasets do not work well in out-of-distribution (OOD) tests, namely when the characteristics of the test data and environment change. Furthermore, practical engineering problems arise when deploying lab-trained models in real scenarios, such as scalability and computation on memory- and resource-limited edge devices, model compression or mixed floating-point calculations, and optimization for embedded devices, etc.
The main goal of this Research Topic is to present new contributions on recent advances in real-world AI and Computer Vision applications. This includes advances in the theory of AI model generalization, the deployment of models, and comparisons between lab-benchmarks and real-world performance. Authors are encouraged to highlight the pros and cons of new methods or established techniques in a specific scenario, focusing on theoretical advances in generalization, and both software and hardware configurations managing trade-offs between performance and feasibility. Contributions may encompass study cases, comparisons of techniques, validation protocols, and practical applications of computer vision and AI.
Research articles are welcome that highlight new approaches in an end-to-end manner, as are review articles surveying state-of-the-art methods in AI generalization and deployment in real-world scenarios. Topics of interest include, but are not limited to:
- Theoretical advances: distribution shifts, domain generalization, generalization, out-of-distribution detection, shortcut learning transfer learning
- Computer Vision and AI tasks: anomaly detection, embedded and edge computing, image and video generation, object detection, classification and tracking, semantic segmentation, remote sensing, scene understanding, visual saliency
- Application scenarios: autonomous vehicles, bio-medical image analysis, art and cultural heritage, healthcare and medicine, marketing and retail, (social) robotics, intelligent surveillance
