Multi-Spectral Vehicle Re-Identification: A Challenge
Authors: Hongchao Li, Chenglong Li, Xianpeng Zhu, Aihua Zheng, Bin Luo11345-11353
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Comprehensive experiments on prevalent networks show that our HAMNet can effectively integrate multi-spectral data for robust vehicle Re-ID in day and night. Our work provides a benchmark dataset for RGB-NIR and RGB-NIR-TIR multi-spectral vehicle Re-ID and a baseline network for both research and industrial communities. |
| Researcher Affiliation | Academia | 1Key Laboratory of Intelligent Computing and Signal Processing of Ministry of Education, School of Computer Science and Technology, Anhui University, Heifei, China 2Key Lab of Industrial Image Processing & Analysis of Anhui Province, Hefei, China |
| Pseudocode | No | The paper describes the network architecture and loss functions textually and with a diagram (Figure 5), but does not provide a formal pseudocode block or algorithm. |
| Open Source Code | Yes | The dataset and baseline codes are available at: https: //github.com/ttaalle/multi-modal-vehicle-Re-ID. |
| Open Datasets | Yes | In this work, we address the RGB and IR vehicle Re-ID problem and contribute a multi-spectral vehicle Re-ID benchmark named RGBN300, including RGB and NIR (Near Infrared) vehicle images of 300 identities from 8 camera views, giving in total 50125 RGB images and 50125 NIR images respectively. The dataset and baseline codes are available at: https: //github.com/ttaalle/multi-modal-vehicle-Re-ID. We are the first time to contribute a standard benchmark dataset RGBN300 to support the study of multi-spectral vehicle Re-ID. We also construct another benchmark datasets with RGB, near infrared and thermal infrared images for related researches and applications. These benchmark datasets will be open to the public for free academic usage. |
| Dataset Splits | Yes | We randomly select 150 vehicles with 25200 image pairs as the training set, while the rest 150 vehicles with 24925 image pairs as the testing set (gallery). We further randomly selected 4985 image pairs from the gallery as the query (probe). In RGBNT100, the training set contains 50 vehicles with 8675 image triples, while the other 50 vehicles with 8575 image pairs for testing/gallery, from which 1715 image triples are randomly selected as the query/probe. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU/CPU models, memory, or processor types used for running experiments. It only implies that models were trained. |
| Software Dependencies | No | The paper mentions using pre-trained networks (ResNet50, DenseNet121, etc.) and the Adam optimizer, but it does not specify version numbers for any of the software components or libraries used for implementation (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | The Adam (Kingma and Ba 2014) optimizer is used with the batch size of 16. We use warmup (Fan et al. 2019) to bootstrap the network, which spent 10 epochs linearly increasing the learning rate from 3.5 10 5 to 3.5 10 4. The learning rate is decayed to 3.5 10 5 and 3.5 10 6 at 40-th epoch and 70-th epoch respectively. Our model is trained in total 120 epochs. The sizes of input images are fixed to 128 256 for Res Net50, Densenet121 and Mobile Net V2, 224 224 for Se Res Net50, and 299 299 for Inception Res Net V3. |