Shape-Adaptive Selection and Measurement for Oriented Object Detection
Authors: Liping Hou, Ke Lu, Jian Xue, Yuqiu Li923-932
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The experimental results on both anchor-free and anchor-based baselines and four publicly available oriented datasets (DOTA, HRSC2016, UCASAOD, and ICDAR2015) demonstrate the effectiveness of the proposed method. |
| Researcher Affiliation | Academia | 1 University of Chinese Academy of Sciences, Beijing 100049, China 2 Peng Cheng Laboratory, Shenzhen 518055, China {houliping17, liyuqiu20}@mails.ucas.ac.cn, {luk, xuejian}@ucas.ac.cn |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The source code of the paper will be available publicly at https: //github.com/houliping/SASM. |
| Open Datasets | Yes | The results of experiments conducted on four typical publicly available datasets containing oriented objects, that is, DOTA (Xia et al. 2018), HRSC2016 (Liu et al. 2017), UCAS-AOD (Zhu et al. 2015), and ICDAR2015 (Karatzas et al. 2015) are summarized to evaluate the effectiveness of the proposed method. |
| Dataset Splits | Yes | DOTA...This dataset contains three subsets, which are the training set (1/2), validation set (1/6), and testing set (1/3)... |
| Hardware Specification | Yes | all experiments were performed using MMDetection-1.1 (Chen et al. 2019) and Py Torch-1.3/1.2 on 2 Titan V GPUs with 11G memory and 4 Tesla V GPUs with 32G memory, while the operating system is Ubuntu 16.04. |
| Software Dependencies | Yes | MMDetection-1.1 (Chen et al. 2019) and Py Torch-1.3/1.2 |
| Experiment Setup | Yes | The framework was trained using the SGD optimizer, where the initial learning rate, momentum, and weight decay were 0.01, 0.9, and 0.0001, respectively. The framework was trained respectively for 12, 36, 120, and 240 epochs on the DOTA, HRSC2016, UCAS-AOD, and ICDAR2015 datasets, respectively. The numbers of the points in a point set in Rep Points and the anchors at each position in S2A-Net were set to 9 and 1, respectively. The weighted parameter ω in (2) was empirically set as 4 on DOTA, UCAS-AOD, and ICDAR2015. Considering that the HRSC2016 dataset contains a large number of elongated ships, ω was set as 14 on it. |