This repository hosts the BeyondSingleObject checkpoints:

The task checkpoint uses the Patch-Interaction Transformer implementation and was jointly fine-tuned on the three released task datasets. Code and training scripts are available at:

Dataset annotations are available at:

Point-cloud preparation instructions and upstream download links are provided in the GitHub repository README.

Expected local layout:

Our latest checkpoint and cleaned datasets further improve multi-object 3D reasoning performance.

The separately released classification checkpoint follows the PointLLM ModelNet40 evaluation protocol. See the repository README for its six one-, two-, and three-object commands.

Example inference:

ModelNet40 classification:

LLM-judged reasoning and delta-caption scores depend on the judge model and prompt configuration. Use the released evaluators and report the exact judge configuration when comparing those metrics. The task and classification checkpoints have different roles and should not be interchanged.

These checkpoints are built with the BeyondSingleObject codebase and use PointLLM-style initialization and data. They may inherit terms from upstream model, code, and dataset components, including PointLLM, Vicuna/Llama, Objaverse/Cap3D, ShapeTalk, Thingi10K, Neural Shape Mating, and ModelNet40. Please check the corresponding upstream licenses before redistribution or commercial use.