基于知识图谱引导的三维视觉问答

毛爱华,陈思宇

3D visual question answering guided by knowledge graph

Aihua MAO,Siyu CHEN

表 2 不同方法在ScanQA不带对象的测试集上的评估指标结果对比

Tab.2 Comparison evaluation metrics for different methods on ScanQA test set without objects

方法 EM@1/% EM@10/% BLEU-1/% BLEU-4/% ROUGE/% METEOR/% CIDEr RandomImage+MCAN[1] 20.82 51.23 26.29 9.66 29.23 11.54 55.64 VoteNet+MCAN[1] 18.15 48.56 29.63 7.10 29.12 11.68 53.34 scanRefer+MCAN(e2e)[1] 19.04 49.70 26.98 7.82 28.61 11.38 53.41 scanQA[1] 20.90 54.11 30.68 10.75 31.09 12.59 60.24 3D-VisTA(scratch) [11] 20.40 51.50 — 8.70 29.60 11.60 55.70 Multi-CLIP[2] 20.71 — 31.22 11.49 31.25 12.80 60.75 本研究 21.10 54.49 33.28 11.94 32.84 13.30 63.41 jin-Context-aware*[4] 21.56 53.89 31.48 15.84 31.79 13.13 63.40 Multi-CLIP(pre-trained)*[2] 21.48 — 32.69 12.87 32.61 13.36 63.20 3D-VisTA*[11] 23.00 53.50 — 11.90 32.80 12.90 62.60 CLIP*[3] 21.37 — 32.70 11.73 32.41 13.28 62.83