Learning High-Precision Bounding Box for Rotated Object Detection via Kullback-Leibler Divergence
Authors: Xue Yang, Xiaojiang Yang, Jirui Yang, Qi Ming, Wentao Wang, Qi Tian, Junchi Yan
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on seven datasets using different detectors show its consistent superiority, and codes are available at https://github.com/yangxue0827/Rotation Detection. Extensive experimental results on seven public datasets and two popular detectors show the effectiveness of our approach, which achieves new state-of-the-art performance for rotation detection. |
| Researcher Affiliation | Collaboration | Xue Yang1 , Xiaojiang Yang1, Jirui Yang2, Qi Ming3, Wentao Wang1, Qi Tian4, Junchi Yan1 1Department of Computer Science and Engineering, Mo E Key Lab of Artificial Intelligence, AI Institute, Shanghai Jiao Tong University 2University of Chinese Academy of Sciences 3Beijing Institute of Technology 4Huawei Inc. |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Experimental results on seven datasets using different detectors show its consistent superiority, and codes are available at https://github.com/yangxue0827/Rotation Detection. The source codes [18] are made public available. |
| Open Datasets | Yes | Our experiments are conducted over a variety of datasets, including three large-scale public datasets for aerial images i.e. DOTA [37], UCAS-AOD [38], HRSC2016 [39], as well as scene text dataset ICDAR2015 [40], MLT [41] and MSRA-TD500 [42]. MS COCO [48] dataset. |
| Dataset Splits | Yes | The proportions of the training set, validation set, and testing set in DOTA-v1.0 are 1/2, 1/6, and 1/3, respectively. we randomly select 1,110 for training and 400 for testing. The training, validation and test set include 436, 181 and 444 images. |
| Hardware Specification | Yes | We use Tensorflow [43] to implement the proposed methods on a server with Tesla V100 and 32G memory. |
| Software Dependencies | No | The paper mentions 'Tensorflow [43]' as the implementation framework but does not specify its version number or any other software dependencies with version information. |
| Experiment Setup | Yes | Weight decay and momentum are set 0.0001 and 0.9, respectively. We employ Momentum Optimizer over 8 GPUs with a total of 8 images per minibatch (1 image per GPU). All the used datasets are trained by 20 epochs in total, and the learning rate is reduced tenfold at 12 epochs and 16 epochs, respectively. The initial learning rate is set to 5e-4. The number of image iterations per epoch for DOTA-v1.0, DOTA-v1.5, DOTA-v2.0, UCAS-AOD, HRSC2016, ICDAR2015, MLT and MSRA-TD500 are 54k, 64k, 80k, 5k, 10k, 10k, 10k and 5k respectively |