RGBD20K: A Large-Scale Benchmark for RGB-D Semantic Segmentation
RGBD20K adds 20,000 RGB-D pairs and 160 classes for semantic segmentation, plus a score-purified fusion method.
RGBD20K is a semantic-segmentation benchmark with 20,000 RGB-D image pairs and 160 fine-grained categories, far more classes than NYUv2 (40) or SUN RGB-D (37). The authors re-check and correct existing labels to reduce annotation noise. They also introduce score-purified fusion (SPF), which they report as state of the art across the benchmarks they evaluate. The dataset is released on GitHub.
- Covers 160 fine-grained classes, versus 40 in NYUv2 and 37 in SUN RGB-D.
- Offers 20,000 RGB-D pairs, a larger training resource than popular benchmarks.
- Labels were re-evaluated and corrected to reduce long-standing annotation noise.
- Score-purified fusion reports state-of-the-art results on the evaluated benchmarks.
Full article171 words · extracted from huggingface.co · click to collapse
In this paper, we propose RGBD20K, a novel dataset for facilitating the development of more robust and general RGB-D semantic segmentation by encompassing abundant categories and high-quality annotations. RGBD20K possesses several attractive properties: (1) Expanded Semantic Space. In particular, it covers 160 fine-grained categories, largely surpassing the category diversity of existing popular RGB-D benchmarks (e.g., NYUv2 with 40 classes and SUN RGB-D with 37 classes). With such enriched semantic coverage, we expect to promote the learning of more generalizable segmentation models. (2) Larger Scale. Compared with current benchmarks, RGBD20K offers 20,000 RGB-D image pairs, providing a substantially larger training resource that benefits the development of more powerful deep models. (3) High-Fidelity Annotation. We perform rigorous re-evaluation and correction of existing labels to resolve long-standing annotation noise, resulting in a clean and reliable ground-truth foundation. Furthermore, we propose a novel score-purified fusion (SPF) method, which achieves state-of-the-art performance across all evaluated benchmarks, demonstrating the effectiveness of our approach in leveraging high-quality multimodal information for RGB-D semantic segmentation. The dataset is here: https://github.com/ShaohuaDong2021/RGBD20K/.
Text extracted automatically; images, tables and formatting may be missing. Original: https://huggingface.co/papers/2609.29028