L6DP: A Benchmark Dataset and Baseline for 6D Object Pose Estimation Under Low-Light Conditions

Vision-based 6D object pose estimation is a fundamental component in visual navigation and robotics applications. However, data collection under low-light conditions remains a major challenge for data-driven pose estimation methods. To address this issue, we introduce LL6DP, a large-scale benchmark dataset comprising extremely low-light RGB-D images with corresponding ground-truth 6D poses, object segmentation masks, and 2D bounding boxes. Based on this dataset, we propose an effective baseline method capable of enhancing low-light images to improve pose estimation performance. We believe that the LL6DP dataset and baseline will contribute significantly to advancing research in 6D pose estimation under low-light scenarios.