Context-Aware Graph Convolution Network for Target Re-identification
Authors: Deyi Ji, Haoran Wang, Hanzhe Hu, Weihao Gan, Wei Wu, Junjie Yan1646-1654
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments show that the proposed method achieves state-of-the-art performance on both person and vehicle reidentification datasets in a plug and play fashion with limited overhead. |
| Researcher Affiliation | Collaboration | 1 Sense Time Research 2 Xidian University 3 Peking University {jideyi, ganweihao, wuwei, yanjunjie}@sensetime.com, wanghaoran@stu.xidian.edu.cn, huhz@pku.edu.cn |
| Pseudocode | No | The paper describes the steps of its method and provides mathematical formulas, but it does not include a clearly labeled 'Pseudocode' or 'Algorithm' block. |
| Open Source Code | No | The paper does not provide an explicit statement about releasing its source code, nor does it include a link to a code repository. |
| Open Datasets | Yes | The Market1501 dataset consists of 32,668 pedestrian image boxes of 1,501 identities, where 12,936 images from 751 identities are used as the training set, 3,368 probe images and 19,732 gallery images of remaining 750 identities are used as the testing set. The Duke MTMC-re ID dataset contains 16,522 images of 702 identities for training, 2,228 probe images of the other 702 identities and 17,661 gallery images of 702 identities for testing. The Ve Ri-776 dataset consists of 49,357 images of 776 distinct vehicles captured by 20 non-overlapping cameras from different viewpoints, resolutions and occlusions. (The paper cites the original sources for these widely recognized public datasets: Zheng et al. 2015, Zheng, Zheng, and Yang 2017, Liu et al. 2016). |
| Dataset Splits | Yes | The Market1501 dataset consists of 32,668 pedestrian image boxes of 1,501 identities, where 12,936 images from 751 identities are used as the training set, 3,368 probe images and 19,732 gallery images of remaining 750 identities are used as the testing set. The Duke MTMC-re ID dataset contains 16,522 images of 702 identities for training, 2,228 probe images of the other 702 identities and 17,661 gallery images of 702 identities for testing. |
| Hardware Specification | No | The paper does not specify any hardware details such as GPU models, CPU types, or other machine specifications used for running the experiments. |
| Software Dependencies | No | The paper mentions using 'Res Net-50' as the backbone and 'Focal loss' but does not specify any software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | We utilize the network proposed in (Luo et al. 2019b) as baseline model and reproduce their results using the same settings, Res Net-50 is used as the backbone. The GCN module in our framework consists of 9 graph convolution layers followed by 3 fully connected layers. We use stochastic gradient descent to optimize our GCN module. The learning rate is 0.01 and the momentum is 0.9. The weight decay rate is 1e-4. The model is trained for 500 epochs with the batch size of 4. Focal loss is adopted as loss function, the α and γ are set to 2 and 0.25. For the hard gallery sampler, we set k1 to 70, k2 to 20 and maximum nodes number k to 100. During the graph construction, only top-8 (i.e. k = 8) nodes are connected to construct a sparse graph. |