VOTSp2026 Starter Kit

Starter kit for the VOTSp2026 Point Tracking Challenge. Includes two ready-to-run tracker wrappers and a guide to submitting your own tracker.

Quickstart

# 1. Install the VOT toolkit and download the dataset
pip install vot-toolkit
vot initialize vots2026/point --workspace ~/votsp_workspace

# 2. Clone this repo
git clone https://github.com/aharley/votsp-starter
cd votsp-starter

# 3. Copy trackers.ini into your workspace and update the paths field
cp trackers.ini ~/votsp_workspace/trackers.ini
# Edit trackers.ini: set paths = /path/to/votsp-starter

# 4. Run a tracker
vot evaluate --workspace ~/votsp_workspace AllTracker

# 5. Pack for submission
vot pack --workspace ~/votsp_workspace AllTracker

Then upload the resulting zip to the evaluation server.

Included trackers

Dependencies (required for both trackers):

pip install torch torchvision opencv-python

AllTracker

AllTracker (Harley et al., ICCV 2025) is the challenge baseline. Weights are downloaded automatically from HuggingFace on first run. Works on both CPU and GPU.

Wrapper: alltracker_tracker.py

CoTracker3

CoTracker3 (Karaev et al., NeurIPS 2024) is another strong point tracker. Weights are downloaded automatically from HuggingFace on first run. Works on both CPU and GPU.

Wrapper: cotracker3_tracker.py

Folder protocol

The VOT toolkit communicates with your tracker by reading and writing text files in a temporary directory. For each sequence, the toolkit writes:

• frames_color.txt — absolute paths to video frames, one per line
• query_N.txt — one file per tracked point:
  • Line 1: frame offset (the query frame index)
  • Line 2: query location as x,y in original image pixel coordinates

Your tracker must write:

• output_N.txt — one line per frame: 0 before the query frame, then x,y from the query frame onward

See example/ for sample files illustrating the format.

Integrating your own tracker

1. Copy one of the wrapper scripts and adapt the tracking logic
2. Add an entry to trackers.ini:

[MyTracker]
command = my_tracker
protocol = folderpython
paths = /path/to/my/tracker/code

3. Run evaluation and pack:

vot evaluate --workspace ~/votsp_workspace MyTracker
vot pack --workspace ~/votsp_workspace MyTracker

4. Register at forms.gle/CWMNkgap61AS2Wuw6 and submit the zip to the evaluation server.

The tracker identifier (the [section] name in trackers.ini) must match the short name you register on the form.

Training restrictions

Training on the following datasets is not allowed: EgoPoints, RoboTAP, AnimalTrack, BADJA, Cell Tracking Challenge, SpaceAnimal, DexYCB, CMU Panoptic, Horse-10, CTMCv1, CroHD, Ant Dataset, Tri-Mouse, I-MuPPET, Schooling-Fish.