3D-LLM: Injecting the 3D World into Large Language Models

Benchmark Model Rank Results
3d-visual-grounding-on-scanrefer-val3D-LLM (BLIP2-flant5)#4Acc@0.25: 30.3
3d-object-captioning-on-objaverse3D-LLM#6GPT-4: 33.42Sentence-BERT: 44.48SimCSE: 43.68Precision: 60.39…
generative-3d-object-classification-on-objaverse3D-LLM#6Objaverse (Average): 45.25Objaverse (I): 49.00Objaverse (C): 41.50
3d-question-answering-3d-qa-on-scanqa-test-w-objects3D-LLM (flamingo)#7Exact Match: 23.2BLEU-1: 32.6BLEU-4: 8.4ROUGE: 34.8METEOR: 13.5…
3d-question-answering-3d-qa-on-scanqa-test-w-objects3D-LLM (BLIP2-flant5)#13Exact Match: 19.1BLEU-1: 38.3BLEU-4: 11.6ROUGE: 35.3…
3d-question-answering-3d-qa-on-scanqa-test-w-objects3D-LLM (BLIP2-opt)#14Exact Match: 19.1BLEU-1: 37.3BLEU-4: 10.7ROUGE: 34.5…
question-answering-on-sqa3d3D-LLM (BLIP2-flant5)#10AnswerExactMatch (Question Answering): 48.1