Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
HFF-Tracker: A Hierarchical Fine-grained Fusion Tracker for Referring Multi-Object Tracking
Authors: Zeyong Zhao, Yanchao Hao, Minghao Zhang, Qingbin Liu, Bo Li, Dianbo Sui, Shizhu He, Xi Chen
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on the Refer-KITTI dataset and the Refer-KITTI-V2 dataset demonstrate that our proposed HFFTracker outperforms other state-of-the-art methods with remarkable margins. |
| Researcher Affiliation | Collaboration | 1Platform and Content Group, Tencent, Beijing 100080, China 2Harbin Institute of Technology, Harbin 150001, China 3The Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences., Beijing 101400, China |
| Pseudocode | No | The paper describes the methodology using prose and mathematical equations (e.g., F vw(i, j, k) = ...), and provides architectural diagrams (Figure 2, 3, 4, 5), but does not contain explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement about releasing its own source code, nor does it provide a link to a code repository. Mentions of 'official open source code and weights' in Table 1 and Table 2 refer to other methods, not the one proposed in this paper. |
| Open Datasets | Yes | Extensive experiments on the Refer-KITTI dataset and the Refer-KITTI-V2 dataset demonstrate that our proposed HFFTracker outperforms other state-of-the-art methods with remarkable margins. Initially, Wu et al. introduce the RMOT task and construct Refer-KITTI dataset(Wu et al. 2023a)...leading to the creation of an improved dataset named Refer-KITTI-v2 (Zhang et al. 2024). |
| Dataset Splits | No | The paper mentions using the Refer-KITTI and Refer-KITTI-V2 datasets and refers to a 'Refer-KITTI testing set' in ablation studies, but it does not provide specific details on how these datasets are split into training, validation, and test sets (e.g., percentages, sample counts, or explicit standard split citations with numerical details). |
| Hardware Specification | Yes | The overall training process is conducted on 8 Nvidia V100 GPUs, with a batch size of 1. |
| Software Dependencies | No | The paper mentions employing models like Res Net50, Ro BERTa, and Deformable DETR, and using the Adam W optimizer. However, it does not provide specific version numbers for any software libraries, frameworks, or programming languages used in the implementation. |
| Experiment Setup | Yes | We employ the Adam W optimizer to train the HFF-Tracker, with a base learning rate of 1e-4. The learning rate of the backbone are set to 1e-5. For the Refer-KITTI dataset, the model is trained for 80 epochs, and for the Refer-KITTI-v2 dataset, it is trained for 70 epochs. The learning rate decays by a factor of 10 at the 40th epoch. The Look-Back-Hard starts at the 5th epoch and ends at the 40th epoch, while the Look Back-Remain only begins after the last epoch of training. During testing, we set the class confident threshold βcls = 0.7 and the reference threshold βref = 0.5 for Refer-KITTI dataset. For the Refer-KITTIV2 dataset, we set and class threshold βcls = 0.6 and reference threshold βref = 0.4. To ensure the repeatability of the experimental results, we set the random seed to 42 for the training stage and 2020 for the inference stage. |