Xiaoye Zhu, Weixin Li, Junan Huo, Bozhong Wang, Jia Zeng, Yi Yang, Cen Chen, Qi Liu†
South China University of Technology
†Corresponding author
Paper | Project Page | Code | Benchmark | Source Assets
CLARE places a clarification gate before 3D tool execution, resolving ambiguity, missing information, and mistaken constraints through targeted dialogue. The release centers on the 3D-Clarify benchmark and a reproducible code-level evaluation pipeline for the paper's single-step and multi-step experiments.
This repository hosts the source media required for physical 3D tool execution on 3D-Clarify. The benchmark JSON and deterministic CR/SR evaluation code are distributed separately in xyzhu1225/3D-Clarify and the CLARE code repository.
CLARE-assets-v1.zip contains 427 image, mesh, and point-cloud files and occupies approximately 443 MiB compressed. The assets support five operation families:
They correspond to the benchmark's 500 single-step tasks and 120 multi-step workflows. Each task is evaluated under ambiguity, missing-information, and mistaken-detail conditions.
Source media is not required for the paper's code-level evaluation. Download it only when enabling the underlying 3D tools.
Download CLARE-assets-v1.zip, then extract it at the root of a CLARE checkout:
unzip CLARE-assets-v1.zip -d /path/to/CLARE/
This creates data/source/, matching the input paths referenced by the released benchmark JSON.
Verify the archive before extraction:
echo "77b5b068f5105e0c95f14d94cd458da52e2187042c8c301e6e7f9076a147fcf9 CLARE-assets-v1.zip" | sha256sum -c -
For benchmark structure, task counts, ground-truth fields, and evaluation commands, see the 3D-Clarify dataset and the dataset guide.
Xiaoye Zhu, Weixin Li, Junan Huo, Bozhong Wang, Jia Zeng, Yi Yang, Cen Chen, Qi Liu†
South China University of Technology
†Corresponding author
Paper | Project Page | Code | Benchmark | Source Assets
CLARE places a clarification gate before 3D tool execution, resolving ambiguity, missing information, and mistaken constraints through targeted dialogue. The release centers on the 3D-Clarify benchmark and a reproducible code-level evaluation pipeline for the paper's single-step and multi-step experiments.
This repository hosts the source media required for physical 3D tool execution on 3D-Clarify. The benchmark JSON and deterministic CR/SR evaluation code are distributed separately in xyzhu1225/3D-Clarify and the CLARE code repository.
CLARE-assets-v1.zip contains 427 image, mesh, and point-cloud files and occupies approximately 443 MiB compressed. The assets support five operation families:
They correspond to the benchmark's 500 single-step tasks and 120 multi-step workflows. Each task is evaluated under ambiguity, missing-information, and mistaken-detail conditions.
Source media is not required for the paper's code-level evaluation. Download it only when enabling the underlying 3D tools.
Download CLARE-assets-v1.zip, then extract it at the root of a CLARE checkout:
unzip CLARE-assets-v1.zip -d /path/to/CLARE/
This creates data/source/, matching the input paths referenced by the released benchmark JSON.
Verify the archive before extraction:
echo "77b5b068f5105e0c95f14d94cd458da52e2187042c8c301e6e7f9076a147fcf9 CLARE-assets-v1.zip" | sha256sum -c -
For benchmark structure, task counts, ground-truth fields, and evaluation commands, see the 3D-Clarify dataset and the dataset guide.