Spatial Transform Decoupling for Oriented Object Detection

Authors: Hongtian Yu, Yunjie Tian, Qixiang Ye, Yunfan Liu

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental STD achieves state-of-the-art performance on the benchmark datasets including DOTA-v1.0 (82.24% m AP) and HRSC2016 (98.55% m AP), which demonstrates the effectiveness of the proposed method.
Researcher Affiliation Academia University of Chinese Academy of Sciences {yuhongtian17, tianyunjie19}@mails.ucas.ac.cn, {qxye, liuyunfan}@ucas.ac.cn
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes Source code is available at https://github.com/yuhongtian17/Spatial-Transform Decoupling.
Open Datasets Yes Experiments are conducted on two commonly-used datasets for oriented object detection, namely DOTA-v1.0 (Xia et al. 2018) and HRSC2016 (Liu et al. 2017).
Dataset Splits Yes The dataset consists of a total of 188,282 individual instances distributed across 15 different classes, and it is partitioned into training, validation, and test sets containing 1,411, 458, and 937 images, respectively. The commonly used training, validation, and test sets consist of 436, 181, and 444 images, respectively.
Hardware Specification Yes All experiments are conducted on 8 A100 GPUs with a batch size of 8.
Software Dependencies No The experimental results are obtained on the MMRotate platform (Zhou et al. 2022).
Experiment Setup Yes The model is trained for 12 epochs on DOTA-v1.0 and 36 epochs on HRSC2016. We adopt the Adam W optimizer (Kingma and Ba 2014) with an initial learning rate of 1e 4/2.5e 4 for DOTA-v1.0/HRSC2016, a weight decay of 0.05, and a layer decay of 0.75/0.90 for Vi T/Hi Vi T.