Do Different Tracking Tasks Require Different Appearance Models?
Authors: Zhongdao Wang, Hengshuang Zhao, Ya-Li Li, Shengjin Wang, Philip Torr, Luca Bertinetto
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In Section 3.1 we perform an extensive evaluation to benchmark a wide variety of off-the-shelf, modern self-supervised models, showing their strengths and weaknesses on all five tasks considered. In Section 3.2 we compare Uni Track (equipped with supervised or unsupervised appearance models) against recent and task-specific tracking methods. |
| Researcher Affiliation | Collaboration | Zhongdao Wang1,2 Hengshuang Zhao3,4 Ya-Li Li1,2 Shengjin Wang1,2 Philip H.S. Torr3 Luca Bertinetto5 1Beijing National Research Center for Information Science and Technology (BNRist) 2Department of Electronic Engineering, Tsinghua University 3Torr Vision Group, University of Oxford 4The University of Hong Kong 5Five AI |
| Pseudocode | No | The paper includes Figure 5, which schematically illustrates the Reconstruction Similarity Metric (RSM) procedure, but it is a diagram and explanatory text, not a formal pseudocode block or algorithm listing. |
| Open Source Code | Yes | https://zhongdao.github.io/Uni Track |
| Open Datasets | Yes | For fair comparison with existing methods, we report results on standard benchmarks with conventional metrics for each task. Please refer to Appendix A for details. |
| Dataset Splits | Yes | Datasets and evaluation metrics. For fair comparison with existing methods, we report results on standard benchmarks with conventional metrics for each task. Please refer to Appendix A for details. [...] (c) Pose Track@Pose Track2018 [2] val split. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU models, CPU types, memory) used for running the experiments. It mentions using ResNet architectures and refers to 'Further implementation details are deferred to Appendix B and C' which is not provided. |
| Software Dependencies | No | The paper mentions 'Py Torch s Model Zoo' but does not specify version numbers for PyTorch or any other software libraries or dependencies. 'Further implementation details are deferred to Appendix B and C'. |
| Experiment Setup | Yes | Implementation details. We use Res Net-18 [25] or Res Net-50 as the default architecture. With Image Net-supervised appearance model, we refer to the Image Net pre-trained weights made available in Py Torch s Model Zoo . To prevent excessive downsampling, we modify the spatial stride of layer3 and layer4 to 1, achieving a total stride of r = 8. We extract features from both layer3 and layer4. |