Saved in:
| Main Author: | |
|---|---|
| Format: | Recurso digital |
| Language: | |
| Published: |
Zenodo
2025
|
| Online Access: | https://doi.org/10.5281/zenodo.15378325 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Table of Contents:
- <p>Object detection is a fundamental task in computer vision with wide-ranging applications in autonomous vehicles, surveillance, healthcare, and robotics [1]. The ability to accurately detect objects in real-time is critical for decision-making in these domains. However, because of increased noise, decreased contrast, and subpar feature extraction, low-light situations provide serious hurdles for traditional object detection algorithms [2]. Due in large part to their dependence on high-quality picture information,<br>traditional Convolutional Neural Networks (CNNs) find it difficult to sustain performance in such circumstances [16]. By utilizing self-attention mechanisms for long-range dependencies, recent developments in Vision Transformers (ViTs) have shown improved performance in object detection [4]. However, ViTs still suffer from performance degradation in low-light scenarios due to suboptimal feature representations. To address this, integrating Adaptive Vision Transformers with Generative Contrastive<br>Learning offers a promising approach. Generative Contrastive Learning enhances the feature extraction process by generating high-quality representations from low-light images, thereby improving model robustness [11]. Despite achieving state-of-the-art performance in object identification, CNN-based models like YOLO, Faster R-CNN, and SSD have major drawbacks when used in low light [12]. This research employs a structured methodology encompassing the Adaptive Vision Transformer implemented<br>with self-attention modules capable of dynamically adjusting to illumination variations, while Generative Contrastive Learning will be incorporated to refine feature extraction and representation learning [15].</p>