Cross-Modality Person Re-identification with Memory-Based Contrastive Embedding
Authors: De Cheng, Xiaolong Wang, Nannan Wang, Zhen Wang, Xiaoyu Wang, Xinbo Gao
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiment results demonstrate the effectiveness of the proposed method, surpassing state-of-the-art works significantly by a large margin on the commonly used VI-Re ID datasets. Experiment Dataset and Evaluation Protocol. Comparison with State-of-the-Art Methods. Ablation Study. |
| Researcher Affiliation | Academia | 1 Xidian University, 2 Zhejiang Lab, 3 University of Science and Technology of China, 4 Chongqing University of Posts and Telecommunications |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide concrete access to source code. There is no link or explicit statement about code availability. |
| Open Datasets | Yes | We evaluate our proposed method on two widely used VI-Re ID datasets: SYSU-MM01 (Wu et al. 2017) and Reg DB (Nguyen et al. 2017). |
| Dataset Splits | No | The paper specifies training and test sets but does not explicitly mention a separate validation set or its split details for hyperparameter tuning. |
| Hardware Specification | Yes | Our model is implemented by Py Torch and trained on a single RTX3090 GPU platform. |
| Software Dependencies | No | The paper mentions "implemented by Py Torch" but does not specify version numbers for PyTorch or any other software dependencies, which is required for reproducibility. |
| Experiment Setup | Yes | During model optimization, we adopt the Adam optimization method, the weight decay is set to 0.0005, the initial learning rate is set to 0.00035 with a warmup strategy, and it is divided by 10 at the 20-th and 40-th epochs. The input images are re-scaled to the size of 384 128 with data augmentation like randomly flipping and erasing used. We train the whole model for 80 epochs overall. For each training step, the batch-size is set to 64, where we randomly sample 8 identities, and each with 4 RGB and 4 infrared images. The momentum update factor in Eq. 3 is set to 0.1 for the modality-agnostic centers and 0.3 for other three modality-aware centers. The hyper-parameters λ1 and λ2 in Eq. 10 is set to 1.2 and 1.0, respectively. All the temperature parameter τ in Eq. 2 5 6 is set to 0.05, and α in Eq. 9 is set to 0.3, empirically. |