Distance-IoU Loss: Faster and Better Learning for Bounding Box Regression
Authors: Zhaohui Zheng, Ping Wang, Wei Liu, Jinze Li, Rongguang Ye, Dongwei Ren12993-13000
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To evaluate our proposed methods, DIo U loss and CIo U loss are incorporated into several state-of-the-art detection algorithms including YOLO v3 (Redmon and Farhadi 2018), SSD (Liu et al. 2016) and Faster R-CNN (Ren et al. 2015), and are evaluated on two popular benchmark datasets PASCAL VOC 2007 (Everingham et al. 2010) and MS COCO 2017 (Lin et al. 2014). |
| Researcher Affiliation | Academia | 1School of Mathematics, Tianjin University, China 2College of Intelligence and Computing, Tianjin University, China 3School of Information Technology and Cyber Security, People s Public Security University of China |
| Pseudocode | Yes | Algorithm 1 Simulation Experiment |
| Open Source Code | Yes | The source code and trained models are available at https://github.com/Zzh-tju/DIo U. |
| Open Datasets | Yes | PASCAL VOC 2007 (Everingham et al. 2010) and MS COCO 2017 (Lin et al. 2014). |
| Dataset Splits | Yes | We use VOC 07+12 (the union of VOC 2007 trainval and VOC 2012 trainval) as training set, containing 16,551 images from 20 classes. And the testing set is VOC 2007 test, which consists of 4,952 images. |
| Hardware Specification | No | No specific hardware details (like GPU/CPU models, processor types, or memory amounts) used for running the experiments were provided. |
| Software Dependencies | No | The paper mentions the use of 'Py Torch implementation of SSD' but does not specify a version number for PyTorch or any other software dependencies. |
| Experiment Setup | Yes | The maximum iteration is set to 50K (YOLO v3) or 120K (SSD). The backbone network is Darknet608 (YOLO v3) or Res Net-50FPN (SSD, Faster R-CNN). For SSD, the weight on regression loss is fixed as 5. For Faster R-CNN, the weight is 12. NMS thresholds are ε = 0.45 for YOLO v3 and SSD, and ε = 0.50 for Faster R-CNN. |