One-Shot Replay: Boosting Incremental Object Detection via Retrospecting One Object
Authors: Dongbao Yang, Yu Zhou, Xiaopeng Hong, Aoting Zhang, Weiping Wang
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental results on VOC2007 and COCO demonstrate that OSR can outperform the state-of-the-art incremental object detection methods without using extra wild data. Experiments Experiment Setup Datasets. The proposed method is evaluated on two benchmark datasets Pascal VOC 2007 and Microsoft COCO 2014. |
| Researcher Affiliation | Academia | Dongbao Yang1,2, Yu Zhou1,2*, Xiaopeng Hong3, Aoting Zhang1,2 , Weiping Wang1 1 Institute of Information Engineering, Chinese Academy of Sciences 2 School of Cyber Security, University of Chinese Academy of Sciences 3 Harbin Institute of Technology {yangdongbao, zhouyu, zhangaoting, wangweiping}@iie.ac.cn, hongxiaopeng@ieee.org |
| Pseudocode | No | The paper describes the method conceptually and with a framework diagram (Figure 2), but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks. |
| Open Source Code | No | The paper does not provide a link to open-source code for the described methodology or explicitly state that the code will be released. |
| Open Datasets | Yes | The proposed method is evaluated on two benchmark datasets Pascal VOC 2007 and Microsoft COCO 2014. Extensive experiments on VOC2007 (Everingham et al. 2010) and COCO (Lin et al. 2014) demonstrate the effectiveness of one-shot replay. |
| Dataset Splits | Yes | VOC2007 has 20 object classes, and we use the trainval subset for training and the test subset for evaluation. For COCO, the 80K images in the training set are used for training, and the minival (the first 5K images from the validation set) split is used for evaluation. |
| Hardware Specification | Yes | The experiments are conducted on a single NVIDIA Ge Force RTX 2080 Ti. |
| Software Dependencies | No | The paper mentions using Faster R-CNN with ResNet-50 as the basic object detector, but does not provide specific version numbers for software dependencies or libraries. |
| Experiment Setup | Yes | The old model is trained for 20 epochs, and the initial learning rate is set to 0.001 (lr = 0.001), and decays every 5 epochs with gamma = 0.1. The momentum is set to 0.9. The new model is trained for 10 epochs with lr = 0.001 and decays after 5 epochs. The confidence and Io U threshold for NMS are set to 0.5 and 0.3 respectively. |