Research data for: Sequence-Level Diagnosis of High-Score Target-Support Inconsistency in Visual Object Tracking

Published: 18 August 2026| Version 2 | DOI: 10.17632/kh7yzbxbgv.2
Contributor:

Description

This dataset contains derived diagnostic summaries, analysis scripts, configuration and command records, and provenance manifests supporting the MCOT and strict-entry analyses reported in the associated article. Raw GOT-10k, LaSOT, UAV123, TrackingNet, and OTB benchmark data are not redistributed and remain available from their official maintainers.Version 2 removes redundant duplicate files and non-scientific workflow artifacts. The scientific data, reported results, analysis code, and configurations are unchanged.

Files

Steps to reproduce

Download and extract the dataset ZIP. Install Python 3 with NumPy and pandas. From the extracted package root, run: python analysis/reproduce_primary_endpoint_from_counts.py The script reads source_data/primary_run_q80_per_sequence_counts.csv and writes verification/primary_endpoint_reproduced.csv. Raw benchmark videos and annotations are not required to reproduce the packaged primary endpoint results.

Categories

Artificial Intelligence, Computer Vision, Machine Learning

Licence