| Dataset
Back to project page Back to collection

Dataset Items: 1 to 20 of 101

A2Z-10M+: Geometric Deep Learning with A-to-Z BRep Annotations for AI-Assisted CAD Modeling and Reverse Engineering

Issue Date Title
Page 1 of 6

Dataset Label Schema

Reference fields for boundary, junction, and face annotations.

Boundary Labels

  • v_vertex_Ids: Indices of scan vertices closest to sampled coedge points on the B-Rep boundary.
  • v_close_loop_Id: ID of the loop that the associated coedge belongs to (i.e., the primary loop).
  • v_close_mate_loop_Id: ID of the loop that is mate to the primary loop via the coedge's mate.
  • v_parentedge_Id: ID of the edge that the coedge is derived from.
  • v_mateface_Id1: ID of the face to which the coedge belongs.
  • v_mateface_Id2: ID of the face corresponding to the coedge's mate.
  • v_edgefeats: Feature dictionary of the associated edge (e.g., curve type such as line, circle, etc.).

Junction Labels

  • j_vertex_Ids: Indices of scan vertices closest to the junctions (start/end points of coedges).
  • j_close_loop_Ids: ID of the loop that the coedge (which contributes this junction) belongs to.
  • j_parentedge_Ids: Edge ID associated with the junction (from the originating coedge).
  • j_mateface_Id1s: Face ID of the junction's originating coedge.
  • j_mateface_Id2s: Face ID of the junction's mate coedge.

Face Labels

  • v_vertex_Ids: Indices of all scan vertices considered for face labeling.
  • v_face_Ids: ID of the closest B-Rep face (by sampled UV-grid points) to each scan vertex.
  • v_face_type: Integer encoding of the face type (e.g., plane, cylinder, sphere), mapped from surface type names.
  • v_face_feat: Feature dictionary for each associated face, containing geometric descriptors like type and possibly curvature info.

OBJ HD

  • vertices: A list of 3D points (x, y, z) that make up the mesh. Each row is one point, stored as float32.
  • faces: A list of triangles that connect the points. Each row has three indices that point into the vertices array, stored as int32.
Resolution note: Chunks 0–69 are 4× upsampled and chunks 70–99 are 16× upsampled.
Open3D example (load NPZ to mesh)
data = np.load("mesh.npz")
                  vertices = data["vertices"]
                  faces = data["faces"]
                  mesh = o3d.geometry.TriangleMesh(
                      o3d.utility.Vector3dVector(vertices),
                      o3d.utility.Vector3iVector(faces)
                  )

Vertices Perturbed

  • v_vertsPerturbed: Final 3D coordinates (x, y, z) after adding small noise to the original vertices, stored as float32.
Note: These are the perturbed coordinates themselves, not per-vertex displacement vectors.

Sketch Files

  • vertices: Boundary-only sketch points in 3D (x, y, z), stored as float32.
  • v_vertex_Ids: Original mesh vertex IDs for each saved boundary-only sketch point, same order as vertices.
  • skip_vertex_Ids: Original mesh vertex IDs that were intentionally skipped or omitted.
Note: Sketch NPZ contains boundary-only vertices; use v_vertex_Ids to map them back to the full mesh.
Note (Boundary/Junction/Face only): All *_Id labels refer to scan "points". Each ID is assigned to a specific point on the boundary, junction, or face. These are per-point labels derived from nearest-neighbor matching to sampled geometry from the B-Rep model.
Title:

A2Z Dataset Chunk 0000

Authors:
Your author list here
Keywords:
A2Z Dataset; CAD Dataset; Geometric Deep Learning
Issue Date:
2026
URI:
https://example.com/a2z
Rights:
🔒
Sign in required Please sign in with your Google account to access the dataset files.

Files in this item

File Description Size Format

Dataset License Agreement

A2Z-10M+ Dataset License Agreement

Last updated: May 2026

The A2Z-10M+ dataset is released by the 3D Vision Group, BITS Pilani, Hyderabad under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license.

By accessing or downloading the Dataset, you agree to the following terms:

You are free to:

Share — copy and redistribute the Dataset in any medium or format.
Adapt — remix, transform, and build upon the Dataset.

Under the following terms:

Attribution (BY) — You must give appropriate credit to the 3D Vision Group, BITS Pilani, Hyderabad, provide a link to the license, and indicate if changes were made. You may do so in any reasonable manner, but not in any way that suggests the licensor endorses you or your use.

NonCommercial (NC) — You may not use the Dataset for commercial purposes. Any commercial use requires prior written consent from the 3D Vision Group.

ShareAlike (SA) — If you remix, transform, or build upon the Dataset, you must distribute your contributions under the same CC BY-NC-SA 4.0 license.

Additional Terms

No Warranty: The Dataset is provided "as is" without warranty of any kind. The 3D Vision Group does not warrant that the Dataset is error-free or complete.

Attribution in Publications: Any publication or presentation using this Dataset must cite the original A2Z-10M+ paper. Please refer to the project page for the recommended citation format.

For the full legal text, see the CC BY-NC-SA 4.0 Legal Code.

By clicking "I Agree" below, you acknowledge that you have read and understood the terms of the CC BY-NC-SA 4.0 license and agree to comply with them.

Your Details (all fields required)

By submitting these details, you confirm that the information provided above is correct to the best of your knowledge. Any misrepresentation or submission of incorrect details may result in the revocation of your license and access to the A2Z-10M+ dataset.