🌟New Function🌟
Now, you can plot the error curve for a single sequence, with different colors representing different trackers. The Y-axis represents tracking error, and the X-axis represents frame count. The plotting method can be seen in draw.py.
This repository contains the necessary tools for RGBT tracking, including datasets（GTOT, RGBT234, LasHeR）, evaluation tools, visualization tools, and results of existing works.
This dataset is derived from the paper "Learning Collaborative Sparse Representation for Grayscale-thermal Tracking" published in 2016 IEEE Transactions on Image Processing (T-IP) by Li et al. from the research group led by Professor Chenglong Li at Anhui University.
The dataset comprises 50 video pairs consisting of sequences of grayscale and thermal infrared images, with each pair having statistical bias. Additionally, the dataset includes ground truth annotations for each frame of the videos and two evaluation metrics.
For more information about the dataset and to download it, please visit gtot.md
This dataset is from the paper "RGB-T Object Tracking: Benchmark and Baseline" published in 2019 Pattern Recognition (PR) by Li et al. from the research group led by Professor Chenglong Li at Anhui University.
Key features of the dataset include:
The dataset comprises 234 pairs of RGB-T video sequences along with their corresponding ground truth annotations (GroundTruth). The video sequences are labeled with 12 attributes. The total number of frames in the dataset is 234,000, with the longest video sequence containing 8,000 frames.
For more information about the dataset and to download it, please visit rgbt234.md
LasHeR consists of 1224 visible and thermal infrared video pairs with more than 730K frame pairs in total. Each frame pair is spatially aligned and manually annotated with a bounding box, making the dataset well and densely annotated. LasHeR is highly diverse capturing from a broad range of object categories, camera viewpoints, scene complexities and environmental factors across seasons, weathers, day and night. Induced by real-world applications, several new challenges are take into consideration in data creation.
For more information about the dataset and to download it, please visit lasher.md
The primary evaluation metrics for RGBT single-object tracking include Precision Rate(PR), Success Rate (SR), and Normalized Precision Rate(NPR).
PR is a measure that describes the Euclidean distance between the center points of the predicted bounding box and the ground truth bounding box. It is generally considered that if the Euclidean distance between the center points of the ground truth bounding box and the predicted bounding box is less than 20 pixels, it indicates accurate localization. However, in the GTOT dataset, this threshold is set to 5 pixels.
The formula for PR is PR = NP/NT,
where NP represents the number of frames where the distance between the tracked algorithm's located target center point and the true target center point is less than a certain threshold, and NT represents the total number of frames.
SR is calculated by measuring the intersection-over-union (IoU) of pixels within the predicted bounding box and the ground truth bounding box. If the IoU exceeds a certain threshold, it is considered a successful tracking.
The formula for SR is SR = NS/NT,
where NS represents the number of frames in which the tracking algorithm successfully tracked the target, and NT represents the total number of frames.
NPR, which stands for Normalized Precision, involves normalizing the predicted bounding box (bbox) and ground truth bbox. By doing so, the performance of the model can be assessed without considering the size of the image resolution.
We have developed a MATLAB-based evaluation toolbox for GTOT, which includes
- methods for calculating PR and SR metrics, generating metric curves, and comparing with existing trackers
- the original tracking result files for the majority of existing trackers
- visualization and analysis methods for tracking
- etc
For detailed instructions on how to use the toolbox and for more information, please refer to the documentation available at GTOT-toolkit. It will provide comprehensive guidance on utilizing the toolbox and accessing additional details about its functionality.
Here are the metrics of existing trackers on the GTOT dataset:
The sources of the trackers are indicated in the following table:
results are provided 🚧building🚧
The official repository for the Lasher dataset provides evaluation tools.
Based on this, we have added some metrics and result files for the latest trackers.
- methods for calculating PR、SR and NPR metrics, generating metric curves, and comparing with existing trackers
- the original tracking result files for the majority of existing trackers
- etc
For detailed instructions on how to use the toolbox and for more information, please refer to the documentation available at Lasher-toolkit. It will provide comprehensive guidance on utilizing the toolbox and accessing additional details about its functionality.
Here are the metrics of existing trackers on the LasHeR dataset:
The sources of the trackers are indicated in the following table:
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent