Class-incremental object detection (CIOD) is a real-world desired capability, requiring an object detector to continuously adapt to new tasks without forgetting learned ones, with the main challenge being catastrophic forgetting. Many methods based on distillation and replay have been proposed to allevi...
Knowledge transferDespite the remarkable performance achieved by DNN-based object detectors, class incremental object detection (CIOD) remains a challenge, in which the network has to learn to detect novel classes sequentially. Catastrophic forgetting is the main problem underlying this difficulty, as ...
Learning Task-Aware Language-Image Representation for Class-Incremental Object Detection 2024 2 CL-DETR (ours) 40.1 Continual Detection Transformer for Incremental Object Detection 2023 3 ERD 34.9 Overcoming Catastrophic Forgetting in Incremental Object Detection via Elastic Response Distillation ...
This repository is about the PyTorch implementation for our AAAI 2022 Paper "Static-Dynamic Co-Teaching for Class-Incremental 3D Object Detection" by Na Zhao and Gim Hee Lee. Deep learning-based approaches have shown remarkable performance in the 3D object detection task. However, they suffer from...
Overcoming catastrophic forgetting in incremental object detection via elastic response distillation. In CVPR, 2022. 2 [13] Raghav Goyal, Samira Ebrahimi Kahou, Vincent Michal- ski, Joanna Materzynska, Susanne Westphal, Heuna Kim, Valentin Haenel, Ingo Fruend, Peter Yianilos, Mo...
YOLOOC: YOLO-based Open-Class Incremental Object Detection with Novel Class Discovery[paper] Beyond the Known: Novel Class Discovery for Open-world Graph Learning[paper] PANDAS: Prototype-based Novel Class Discovery and Detection[paper][code] ...
Incremental Learning: Most studies in incremental learn- ing have focused on object detection and classification problems [5, 30, 41, 43, 47]. Some of these works use replay-based approaches, which store samples from pre- vious tas...
Incremental learning of object detectors Shmelkov等人将LwF应用于目标检测任务。但是,它们的框架只能应用于外部计算提案的对象检测器,例如Fast R-CNN[13]。 实验表明,该方法适用于更高效的single-shot目标检测体系结构,如RetinaNet[30]。 Exploiting external data ...
Set the checkpointRunId property: The id of a previous run that has a pretrained checkpoint for incremental training. ImageModelSettingsObjectDetection withDistributed(Boolean distributed) Set the distributed property: Whether to use distributed training. ImageModelSettingsObjectDetection ...
Furthermore, we introduce the Few-shot Class-incremental Classification (FSCIC) methods from data-based, structure-based, and optimization-based approaches and the Few-shot Class-incremental Object Detection (FSCIOD) methods from anchor-free and anchor-based approaches. Beyond these, we present ...