schemapose_data / README.md
anonymouspaper2026's picture
Update README.md
ea27e48 verified
metadata
license: other
pretty_name: SchemaPose Anonymous Synthetic Data
tags:
  - synthetic-data
  - pose-estimation
  - schema-annotation

Anonymous SchemaPose Synthetic Data

This repository provides anonymous supplementary data for the submitted paper on SchemaPose. The data are released for review purposes and are intended to support reproducibility of the schema-driven data generation and training pipeline.

Files

  • AP_data.zip
    SchemaData used for training SchemaPose. It contains synthetic RGB images and corresponding annotations generated by the schema-driven data generation pipeline.

  • objs.zip
    Generated object instances. Each category contains 10,000 procedurally generated object models with intra-category shape variation.

  • real275genes.zip
    Schema annotations provided for the REAL275 dataset in our experiments.

  • additional_videos.zip
    Visualization results for REAL275 test scenes.

Data Description

The synthetic data are generated from parametric category schemas. Each object instance is created by sampling discrete structural choices and continuous geometric parameters, and is therefore associated with accurate schema annotations by construction. The generated object models are then rendered into RGB scenes with corresponding object-level pose annotations.

The released data include:

  • synthetic RGB data and annotations;
  • generated object models;
  • schema labels for generated objects;
  • schema annotations for REAL275.

Intended Use

These files are intended for reproducing the experiments and ablations reported in the submitted paper. In particular, they support:

  • training with schema supervision;
  • mixed training with synthetic and real data;
  • evaluation of schema-guided pose estimation;
  • ablation studies on schema-conditioned decoding.

Anonymity

This repository has been anonymized for peer review. File names, metadata, and documentation are intended to avoid revealing author identity or affiliation.

Notes

The full implementation and training scripts are provided separately in the supplementary code package. Please refer to the code package for environment setup, data preparation, training commands, and evaluation commands.