Dimensions overall score 7.0

• Industrial3D is the largest publicly available terrestrial LiDAR dataset for industrial MEP scene understanding, containing 612 million labelled points.result 95%
• Industrial3D is 6.6 times larger than the closest comparable MEP dataset (Jing et al., 2024).result 95%
• The best supervised method evaluated on Industrial3D achieves 55.74% mIoU, while zero-shot Point-SAM reaches only 15.79%, revealing a 39.95 percentage-point performance gap.result 95%
• Industrial3D exhibits extreme class imbalance with a ratio of 215:1, which is 3.5 times more severe than the S3DIS dataset.technical 90%
• Existing architectural benchmarks like S3DIS and ScanNet create a double domain gap (scene type and sensor modality) when applied to TLS acquisitions of industrial plants.limitation 85%
• The performance gap on Industrial3D originates from a dual crisis of statistical rarity and geometric ambiguity where tail-class points share cylindrical primitives with head-class pipes.technical 85%
• Industrial3D enables studies that were previously infeasible, including 0.1% sparse-supervision experiments and cross-area generalisation across multiple facilities.method 80%
• The dataset comprises 13 areas across 13 facility types, covering over 15,000 m², with data acquired using a Leica BLK360 terrestrial laser scanner.method 90%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card