DRGBT-1K: A Large-scale High-quality Benchmark for Dynamic RGBT Tracking

2026-07-23 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

DRGBT-1K is a new large-scale, high-quality benchmark designed for Dynamic RGBT (DRGBT) tracking, addressing the limitations of existing datasets in evaluating tracker robustness under real dynamic modality and cross-platform variations. It features 1,045 sequences and 795K RGBT frame pairs, captured entirely in real-world scenarios using UAVs and handheld RGBT devices. The benchmark provides comprehensive annotations, including dense bounding boxes, 24 target categories, 15 challenge attributes, and frame-level modality and platform labels. It also includes an unaligned version of DRGBT-1K and UGVT-1K for UAV-ground collaborative tracking. Evaluation of 20 representative multimodal tracking methods revealed significant performance degradation, particularly under platform variations.

Key takeaway

For Machine Learning Engineers developing robust object tracking systems, DRGBT-1K highlights the critical need for solutions that handle dynamic modality changes and cross-platform transitions. Existing RGBT trackers struggle with real-world viewpoint shifts and appearance discontinuities. You should prioritize developing algorithms that effectively integrate multimodal cues while adapting to severe scale and aspect ratio variations inherent in dynamic environments.

Key insights

Robust dynamic RGBT tracking necessitates benchmarks that accurately reflect real-world modality and platform variations.

Principles

• Real-world data is crucial for dynamic tracking benchmarks.
• Platform transitions introduce significant target appearance variations.
• Unified evaluation protocols enable fair tracker comparison.

Method

DRGBT-1K was constructed by collecting 1,045 real-world sequences using UAVs and ground cameras, followed by temporal/spatial alignment and dual-view collaborative annotation for bounding boxes, attributes, categories, and modality/platform labels.

In practice

• Evaluate trackers on combined modality and platform variations.
• Develop methods robust to viewpoint and scale changes.
• Utilize unaligned data for practical sensor misalignment research.

Topics

• Dynamic RGBT Tracking
• Multimodal Tracking
• Object Tracking Benchmarks
• UAV-Ground Tracking
• Computer Vision Datasets
• Sensor Fusion

Code references

• zhaodongAH/DRGBT-1K

Best for: Research Scientist, AI Scientist, Computer Vision Engineer, Machine Learning Engineer

Related on AIssential

• DRGBT-1K: A Large-scale High-quality Benchmark for Dynamic RGBT Tracking — DRGBT-1K provides a robust benchmark for evaluating dynamic RGBT tracking under real-world modality and platform variations.
• Active Adversarial Perturbation-driven Associative Memory Retrieval for RGB-Event Visual Object Tracking — APRTrack improves RGB-Event tracking robustness by simulating real-world degradations and using a novel historical information retrieval method.
• DynTrace: Tracking Dynamic Object Evidence for 4D Spatio-Temporal Reasoning in MLLMs — DynTrace enhances MLLMs' 4D spatio-temporal reasoning by continuously tracking dynamic object evidence using geometry-informed visual priors.
• 🎂REMIND: long-term MOT re-ID🎂 👉REMIND by CVAR-UPM is a novel online tracker designed for... — REMIND enables long-term multi-object re-identification of indoor objects using only monocular RGB, without camera pose or depth.
• Event-Adaptive State Transition and Gated Fusion for RGB-Event Object Tracking — MambaTrack dynamically adapts state transitions and fuses RGB-event data based on event density for robust object tracking.
• DUALVISION: RGB-Infrared Multimodal Large Language Models for Robust Visual Reasoning — Integrating infrared and RGB data via localized cross-attention significantly boosts MLLM robustness in degraded visual conditions.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.