MM-Tracker: Motion Mamba for UAV-platform Multiple Object Tracking
Authors: Mufeng Yao, Jinlong Peng, Qingdong He, Bo Peng, Hao Chen, Mingmin Chi, Chao Liu, Jon Atli Benediktsson
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Based on the Motion Mamba module and motion margin loss, our proposed MM-Tracker surpasses the state-of-the-art in two widely open-source UAV-MOT datasets. ... Experiments Datasets and Metrics. ... Ablation study. ... Comparison with state-of-the-art. |
| Researcher Affiliation | Collaboration | 1School of Computer Science, Shanghai Key Laboratory of Data Science, Fudan University 2Tencent Youtu Lab 3Shanghai Ocean University 4University of Iceland |
| Pseudocode | No | The paper describes the proposed methods, Motion Mamba Module and Motion Margin Loss, using mathematical equations, figures, and textual descriptions, but it does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code https://github.com/Yao Mufeng/MMTracker |
| Open Datasets | Yes | We conduct comparative experiments on the Visdrone (Zhu et al. 2020) and UAVDT (Du et al. 2018) datasets. These two datasets are open-sourced multi-class multi-object tracking datasets, and both are collected from the perspective of UAVs. |
| Dataset Splits | Yes | The Visdrone dataset consists of a training set (56 sequences), validation set (7 sequences), test-dev set (7 sequences), and test-challenge set (6 sequences). ... The UAVDT dataset consists of a training set (30 sequences), and a test set (20 sequences)... |
| Hardware Specification | Yes | the training and testing are completed on a single 2080TI graphic card. |
| Software Dependencies | No | The paper mentions using YOLOX-S as the base object detector and stochastic gradient descent for optimization, but it does not specify software dependencies with version numbers (e.g., Python 3.x, PyTorch 1.x, CUDA 11.x). |
| Experiment Setup | Yes | We use the YOLOX-S (Ge et al. 2021) model as the base object detector on both datasets, the input image size is 1088 608. We use the stochastic gradient descent method (Bottou 2010) to optimize the detector, the learning rate is set to 0.0001, the batch size is set to 8, each dataset is trained for 10 epochs... |