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 ...
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...
They further improved the detection accuracy by modifying the network structure and adding normalization to the predicted parameters. Dong et al. (2022) first introduced the DETR object detector (Zhu et al., 2020) into few-shot object detection and proposed Incremental-DETR. They still followed ...
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...
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 ...
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] ...
The fit function fits a configured one-class support vector machine (SVM) model for incremental anomaly detection (incrementalOneClassSVM object) to streaming data.
some sort of feature selection may be required to keep the signal to noise ratio high and keep computational costs reasonable. Fourth, judging algorithms solely on prediction performance on a holdout dataset is taking a rather narrow view of performance. A very tiny incremental increase in predicti...
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...
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 ...