IJCV 2024 : International Journal of Computer Vision

Visual recognition is a critical task in computer vision, which has gained sig­nificant attention in recent years due to its numerous applications in various fields, including image classification, object detection, semantic segmentation, and instance retrieval for autonomous driving and intelligent surveillance. Over the past decade, deep learning techniques and large-scale datasets have con­tributed to remarkable advancements in the performance of visual recognition systems. However, existing visual models are often limited by their closed-world assumptions, where all possible classes and domains are known in advance, and all data is given at once. Such assumptions are not applicable in practical sce­narios, where novel or previously unseen classes and domains can arise, and data may be continually coming or decentralized due to data privacy concerns. One example of such a scenario is an autonomous vehicle encounters new traffic patterns, a medical AI system detects new diseases, or an intelligent system en­counters a criminal wearing a new type of disguise or clothing. Another example is when a healthcare organization collects patient data from multiple hospitals, and the data may be decentralized and continually coming in due to privacy concerns. Moreover, the challenge of multi-modality poses another obstacle, as visual recognition systems must be capable of integrating and handling data from multiple sources, such as images, videos, texts, and 3D models. Therefore, new methods and techniques are needed to address the challenges of open-world visual recognition and enable visual recognition systems to perform effectively in practical scenarios.

Aims & Scope

This special issue invites innovative research papers that aim to address these challenges and propose novel techniques for open-world visual recognition. Potential topics of interest include, but are not limited to:

Novel Class Discovery, where the goal is to discover underlying semantic clusters for unlabeled data including unseen classes
Open-Set Semi-Supervised Learning, where the system should learn from both labeled and unlabeled data, and distinguish between known and unknown classes during testing
Open Vocabulary Visual Learning, where the goal is to train a model that can recognize a broader range of visual concepts, including new and rare ones, using a limited pre-defined set of categories
Out-of-Distribution Detection, where the goal is to distinguish between in-distribution and out-of-distribution samples
Robust/Adversarial Learning, where the goal is to improve the model’s robustness to distributional shifts, such as adversarial perturbations and corruptions
Open-World Domain Adaptation, where the goal is to adapt/generalize the model to target domains under open-world scenarios, such as universal, source-free, test-time domain adaptation, and domain generalization

Moreover, this special issue also welcomes papers that focus on

Developing new techniques for continual learning, federated learning and multi-modality learning in the context of open-world visual recognition
Building datasets and benchmarks for facilitating the study of open-world visual recognition

We encourage submissions that cover a broad range of visual recognition tasks, including but not limited to image classification, object detection, se­mantic segmentation, action recognition and pose estimation.

This special issue will provide a platform for researchers to share their latest findings and contribute to the advancement of open-world visual recognition. The contributions in this special issue could significantly benefit society by en­abling more robust and reliable visual recognition systems, enhancing public safety and security, improving healthcare, and increasing the efficiency of indus­trial and commercial applications.