PVCap: Towards Accurate 3D Dense Captioning via PseudoCap and VoxelCapNet

Benchmark Model Rank Results
3d-dense-captioning-on-nr3dPVCap-Swin3D (MLE)–CIDEr: 53.31BLEU-4: 30.14METEOR: 27.61ROUGE-L: 59.98
3d-dense-captioning-on-scanrefer-datasetPVCap-SparseUNet (MLE)–CIDEr: 72.21BLEU-4: 40.95METEOR: 28.83ROUGE-L: 62.88
3d-dense-captioning-on-scanrefer-datasetPVCap-Swin3D (MLE)–CIDEr: 77.33BLEU-4: 40.78METEOR: 29.39ROUGE-L: 62.73