PointLLM: Empowering Large Language Models to Understand Point Clouds

Benchmark Model Rank Results
3d-object-captioning-on-objaverse-1PointLLM-13B V1.2#3GPT-4: 48.15Sentence-BERT: 47.91SimCSE: 49.12Precision: 78.75
3d-object-captioning-on-objaverse-1PointLLM-7B V1.2#5GPT-4: 44.85Sentence-BERT: 47.47SimCSE: 48.55Precision: 82.14
3d-question-answering-3d-qa-on-3d-mm-vetPointLLM-13B v1.2#3Overall Accuracy: 46.6
3d-question-answering-3d-qa-on-3d-mm-vetPointLLM-7B v1.2#4Overall Accuracy: 41.2
generative-3d-object-classification-on-1PointLLM-13B v1.2#3Objaverse (Average): 54.00Objaverse (I): 56.50Objaverse (C): 51.50
generative-3d-object-classification-on-1PointLLM-7B v1.2#5Objaverse (Average): 53.00Objaverse (I): 55.00Objaverse (C): 51.00
generative-3d-object-classification-on-2PointLLM-13B v1.2#4ModelNet40 (Average): 52.78ModelNet40 (I): 53.00ModelNet40 (C): 52.55
generative-3d-object-classification-on-2PointLLM-7B v1.2#5ModelNet40 (Average): 52.63ModelNet40 (I): 53.44ModelNet40 (C): 51.82