RGBD1K: A Large-Scale Dataset and Benchmark for RGB-D Object Tracking
Authors: Xue-Feng Zhu, Tianyang Xu, Zhangyong Tang, Zucheng Wu, Haodong Liu, Xiao Yang, Xiao-Jun Wu, Josef Kittler
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The results, of extensive experiments using the SPT tracker demonstrate the potential of the RGBD1K dataset to improve the performance of RGBD tracking, inspiring future developments of effective tracker designs. |
| Researcher Affiliation | Academia | 1School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi, Jiangsu, P.R. China 2Centre for Vision, Speech and Signal Processing, University of Surrey, Guildford, GU2 7XH, UK |
| Pseudocode | No | The paper does not include pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | Yes | The dataset and codes will be available on the project homepage: https://github.com/xuefeng-zhu5/RGBD1K. |
| Open Datasets | Yes | To address the dataset deficiency issue, a new RGB-D dataset named RGBD1K is released in this paper. The RGBD1K contains 1,050 sequences with about 2.5M frames in total. ... The dataset and codes will be available on the project homepage: https://github.com/xuefeng-zhu5/RGBD1K. |
| Dataset Splits | Yes | RGBD1K contains 1,050 sequences with about 2.5M frames in total. Of these, 1,000 videos are reserved for training and 50 videos for testing. For the training videos, ... only the first 600 frames of each video are annotated. Therefore, 600,000 annotated frames of RGBD1K can be utilized for supervised learning of deep RGB-D tracking methods. |
| Hardware Specification | Yes | The proposed SPT tracker is trained and evaluated with an Intel i9-CPU and one NVIDIA Ge Force RTX 3090 GPU. |
| Software Dependencies | No | The paper mentions using ResNet-50 as a backbone and initializing weights from the STARK-S model, but it does not specify version numbers for any software libraries, frameworks (e.g., PyTorch, TensorFlow), or programming languages. |
| Experiment Setup | Yes | The training and test parameters are set the same as Stark, except for the learning rate and training epoch number. The learning rate is set as 10 5 and the total epoch number is 250. As to the backbone, transformer encoder A, B, transformer decoder and box prediction head of SPT, we initialize their weights by using the weights of corresponding components of the officially published STARK-S model. |