Abstract:Domain adaptation is an inspiring solution to the misalignment issue of day/night image features for nighttime UAV tracking. However, the one-step adaptation paradigm is inadequate in addressing the prevalent difficulties posed by low-resolution (LR) objects when viewed from the UAVs at night, owing to the blurry edge contour and limited detail information. Moreover, these approaches struggle to perceive LR objects disturbed by nighttime noise. To address these challenges, this work proposes a novel progressive alignment paradigm, named domain-aware diffusion model (DaDiff), aligning nighttime LR object features to the daytime by virtue of progressive and stable generations. The proposed DaDiff includes an alignment encoder to enhance the detail information of nighttime LR objects, a tracking-oriented layer designed to achieve close collaboration with tracking tasks, and a successive distribution discriminator presented to distinguish different feature distributions at each diffusion timestep successively. Furthermore, an elaborate nighttime UAV tracking benchmark is constructed for LR objects, namely NUT-LR, consisting of 100 annotated sequences. Exhaustive experiments have demonstrated the robustness and feature alignment ability of the proposed DaDiff. The source code and video demo are available at <a class="link-external link-https" href="https://github.com/vision4robotics/DaDiff" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

This paper attempts to solve the feature alignment problem in unmanned aerial vehicle (UAV) tracking at night. Specifically, the existing UAV tracking methods perform well in daytime scenes, but at night, due to insufficient illumination, low signal - to - noise ratio and decreased contrast, etc., the image feature distributions are very different, making these methods difficult to work effectively. Especially for low - resolution (LR) target objects, the blurred edges and limited detail information of night images further exacerbate this problem. ### Main problem summary: 1. **Feature distribution difference**: There are significant differences in image features between day and night, resulting in poor performance of existing trackers at night. 2. **Challenges of low - resolution objects**: Low - resolution objects in night images are difficult to be accurately identified and tracked due to insufficient detail information and noise interference. 3. **Instability of one - step adaptation paradigm**: The existing one - step domain adaptation methods are unstable when dealing with low - resolution objects and are difficult to effectively align the feature distributions. ### Solutions: To solve the above problems, the paper proposes a new progressive alignment paradigm - Domain - aware Diffusion Model (DaDiff). This model improves the robustness and accuracy of UAV tracking at night in the following ways: - **Domain - aware Diffusion Model (DaDiff)**: Gradually enhances the detail information of low - resolution objects at night through a multi - step diffusion process, and steadily narrows the gap between day and night feature distributions. - **Alignment Encoder**: Used to enhance the detail information of low - resolution objects at night. - **Tracking - oriented Layer**: Ensures close cooperation with the tracking task and integrates effective domain - aware information. - **Successive Distribution Discriminator**: Used to distinguish different feature distributions at each diffusion time step and ensure the stability of the alignment process. In addition, the paper also constructs a new benchmark dataset named NUT - LR, which is specifically used to evaluate the tracking performance of low - resolution objects at night. This dataset contains 100 annotated sequences, covering low - resolution objects in various practical application scenarios. ### Conclusion: By introducing the diffusion model and combining the progressive alignment strategy, DaDiff can align the features of low - resolution objects at night more stably and controllably, thereby significantly improving the performance of UAV tracking at night. The experimental results show that DaDiff outperforms existing methods on multiple benchmark datasets, especially when dealing with low - resolution objects.

DaDiff: Domain-aware Diffusion Model for Nighttime UAV Tracking

Prompt-Driven Temporal Domain Adaptation for Nighttime UAV Tracking

Unsupervised Domain Adaptation for Nighttime Aerial Tracking

SAM-DA: UAV Tracks Anything at Night with SAM-Powered Domain Adaptation

Enhancing Nighttime UAV Tracking with Light Distribution Suppression

Mutual-Learning Knowledge Distillation for Nighttime UAV Tracking

Unsupervised Nighttime Object Tracking Based on Transformer and Domain Adaptation Fusion Network

ADTrack: Target-Aware Dual Filter Learning for Real-Time Anti-Dark UAV Tracking

All-Day Object Tracking for Unmanned Aerial Vehicle

Tracker Meets Night: A Transformer Enhancer for UAV Tracking

DCPT: Darkness Clue-Prompted Tracking in Nighttime UAVs

MambaNUT: Nighttime UAV Tracking via Mamba and Adaptive Curriculum Learning

AutoTrack: Towards High-Performance Visual Tracking for UAV With Automatic Spatio-Temporal Regularization

Scale-Aware Domain Adaptation for Robust UAV Tracking

Conditional Generative Denoiser for Nighttime UAV Tracking

Towards Discriminative Representations with Contrastive Instances for Real-Time UAV Tracking

Enhancing UAV tracking: a focus on discriminative representations using contrastive instances

SGDViT: Saliency-Guided Dynamic Vision Transformer for UAV Tracking

Learning Aberrance Repressed Correlation Filters for Real-Time UAV Tracking

Learning Motion Blur Robust Vision Transformers with Dynamic Early Exit for Real-Time UAV Tracking

Light the Night: A Multi-Condition Diffusion Framework for Unpaired Low-Light Enhancement in Autonomous Driving