---
license: cc-by-4.0
task_categories:
- time-series-forecasting
tags:
- G-code,
- 3D_printing
- Time-Series
- Print_Duration
- Regression
- Extrinsic_Regression
- Multivariate
pretty_name: >-
  3DTime: A Large Dataset of Multi-Annotated Multivariate Time-Series for
  3D-printing Duration
size_categories:
- 10M<n<100M
---

# 3DTime: A Large Dataset of Multi-Annotated Multivariate Time-Series for 3D-printing Duration

3DTime is a paper dataset currently under review at the AAAI-26 conference, for the main technical track.

We could not find a way to anonymously publish the full 1.2 TB dataset, hence this smaller version.

## Dataset content

The original dataset contains:
- 9,930 3D models, each sliced 4 times
- 39,720 annotated G-code files (compressed)
- 39,720 binary files
- A total of 12,442,224,222 G-code instructions

This smaller version contains:
- 100 3D models (~1% of the full dataset), each sliced 4 times
- 400 annotated G-code files (compressed)
- 400 binary files
- A total of 64,394,143 G-code instructions (~0.5% of the full dataset)

## Data loader for PyTorch

Due to the peculiar data format of this dataset (each G-code file has a specific length, some being over several million instructions long), we provide a custom made data loader designed for PyTorch utilization, instead of the standard Hugging Face automatic loader.

The code can be found in the complementary material of the submission.

This repository contains all of the code required for the dataset generation, statistic analysis, and model trainings that were presented in the paper.