I noticed that the paper mentions the model is capable of point cloud description and point cloud content-based question answering (QA). However, in the task folder of the current codebase, only the code for the caption3d task is available. I made minor modifications to it using Codex to enable it to output questions and options from prompts, but the model fails to follow my instructions to answer the questions—instead, it still only generates descriptions of the point cloud. Could you please provide the official code for the QA task, or suggest a solution to this issue? The checkpoint I am using is: https://huggingface.co/alexzyqi/GPT4Point/blob/main/gpt4point_pretrain_stage2_opt2.7b.pth
I noticed that the paper mentions the model is capable of point cloud description and point cloud content-based question answering (QA). However, in the task folder of the current codebase, only the code for the caption3d task is available. I made minor modifications to it using Codex to enable it to output questions and options from prompts, but the model fails to follow my instructions to answer the questions—instead, it still only generates descriptions of the point cloud. Could you please provide the official code for the QA task, or suggest a solution to this issue? The checkpoint I am using is: https://huggingface.co/alexzyqi/GPT4Point/blob/main/gpt4point_pretrain_stage2_opt2.7b.pth