In real-world applications, an object detector often encounters object
instances from new classes and needs to accommodate them effectively. Previous
work formulated this critical problem as incremental object detection (IOD),
which assumes the object instances of new classes to be fully annotated in
incremental data. However, as supervisory signals are usually rare and
expensive, the supervised IOD may not be practical for implementation. In this
work, we consider a more realistic setting named semi-supervised IOD (SSIOD),
where the object detector needs to learn new classes incrementally from a few
labelled data and massive unlabelled data without catastrophic forgetting of
old classes. A commonly-used strategy for supervised IOD is to encourage the
current model (as a student) to mimic the behavior of the old model (as a
teacher), but it generally fails in SSIOD because a dominant number of object
instances from old and new classes are coexisting and unlabelled, with the
teacher only recognizing a fraction of them. Observing that learning only the
classes of interest tends to preclude detection of other classes, we propose to
bridge the coexistence of unlabelled classes by constructing two teacher models
respectively for old and new classes, and using the concatenation of their
predictions to instruct the student. This approach is referred to as
DualTeacher, which can serve as a strong baseline for SSIOD with limited
resource overhead and no extra hyperparameters. We build various benchmarks for
SSIOD and perform extensive experiments to demonstrate the superiority of our
approach (e.g., the performance lead is up to 18.28 AP on MS-COCO). Our code is
available at url{https://github.com/chuxiuhong/DualTeacher}.
In this article, the authors discuss the problem of object detection in the context of new classes being introduced. They introduce the concept of incremental object detection (IOD) which assumes that object instances of new classes are fully annotated in incremental data. However, they acknowledge that this approach may not be practical due to the rarity and expense of supervisory signals.
To address this issue, the authors propose a more realistic setting called semi-supervised IOD (SSIOD), where the object detector needs to learn new classes incrementally from a few labelled data and massive unlabelled data without forgetting the old classes. They propose a strategy called DualTeacher, which involves constructing two teacher models for old and new classes, respectively, and using their predictions to instruct the student model. This approach aims to bridge the coexistence of unlabelled classes and provide guidance to the object detector.
The authors highlight the multi-disciplinary nature of this concept, as it combines ideas from computer vision (object detection), machine learning (incremental learning), and semi-supervised learning (learning from limited labelled data and abundant unlabelled data). By leveraging the strengths of different fields, they aim to improve the performance of SSIOD.
To evaluate the effectiveness of their approach, the authors build various benchmarks for SSIOD and conduct extensive experiments. They demonstrate that their DualTeacher approach outperforms existing methods, with a performance lead of up to 18.28 AP on MS-COCO dataset.
This research has practical implications for solving real-world problems in object detection. By addressing the limitations of supervised IOD and proposing an efficient approach for semi-supervised learning, the authors contribute to the development of more robust and adaptable object detection systems.