End-to-End Thorough Body Perception for Person Search
Authors: Kun Tian, Houjing Huang, Yun Ye, Shiyu Li, Jinbin Lin, Guan Huang12079-12086
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments with ablation analysis show the effectiveness of our proposed end-to-end multi-task model, and we demonstrate its superiority over the state-of-the-art methods on two benchmark datasets including CUHK-SYSU and PRW. |
| Researcher Affiliation | Collaboration | 1Horizon Robotics, 2Institute of Automation, Chinese Academy of Sciences {kun.tian, yun.ye, shiyu.li, jinbin.lin}@horizon.ai, houjing.huang@nlpr.ia.ac.cn, huangguan13@mails.ucas.ac.cn |
| Pseudocode | Yes | Algorithm 1: Part-aligned feature learning algorithm |
| Open Source Code | No | No explicit statement or link providing access to the open-source code for the described methodology was found in the paper. |
| Open Datasets | Yes | CUHK-SYSU contains 11,206 images with 5,532 labeled person IDs for training, 2,900 queries and a total of 6,978 gallery images for testing. PRW consists of 11,816 annotated video frames, 933 labeled identities, and 43,110 person bounding boxes. Following the definition in COCO dataset (Lin et al. 2014). |
| Dataset Splits | No | The paper describes training and test sets for CUHK-SYSU and PRW datasets with specific counts, but does not explicitly mention a separate validation set or a clear train/validation/test split for its own experiments. |
| Hardware Specification | Yes | each mini-batch has 4 or 3 images on one GTX 1080Ti due to the memory limitation. |
| Software Dependencies | No | The model is trained under the MXNet framework, but no specific version number for MXNet or other key software dependencies is provided. |
| Experiment Setup | Yes | The network is optimized for 23 epochs using mini-batch stochastic gradient descent with a weight decay of 0.00004 and a momentum of 0.9. We adopt the warmup strategy (Goyal et al. 2017) to change the learning rate for the first 5 epochs and divide it by 10 at 19-th and 22th epoch. The temperature scalar τ in Equation 2, 3 is set to 0.05. For CUHK-SYSU and PRW, the size of the Circular Queue is set to 5,000 and 500 respectively. The scale of the training and testing images is fixed to 600 1000 pixels unless otherwise noted and the training sample are also horizontally flipped in random. |