The CC3D-Ops dataset contains 37k+ B-Reps with the corresponding per-face CAD operation type and step annotations. These labels were extracted using the Solidworks API. The B-Reps and their corresponding annotations constitute an extension of the CC3D dataset [1]. While the Fusion360 dataset [2] contains a similar number of B-Reps (35k+) with the corresponding CAD operation type labels, it does not provide CAD operation step labels, and it includes relatively simple CAD models. The proposed CC3D-Ops dataset comes with more complex models that are closer to real-world industrial challenges.
[1] Pvdeconv: Point-voxel deconvolution for autoencoding cad construction in 3d, Cherenkova, Kseniya, Djamila Aouada, and Gleb Gusev, 2020 IEEE International Conference on Image Processing (ICIP).
[2] Fusion 360 gallery: A dataset and envi- ronment for programmatic cad construction from human de- sign sequences, Karl D. D. Willis, Yewen Pu, Jieliang Luo, Hang Chu, Tao Du, Joseph G. Lambourne, Armando Solar-Lezama, and Wojciech Matusik, ACM Trans. Graph., 40(4), 2021.