SortScrew

SortScrew is a dataset for screw classification. We collected 560 images of 6 types of screws.

Citation

@misc{fu2026sortscrewsdatasetbaselinerealtime,
      title={SortScrews: A Dataset and Baseline for Real-time Screw Classification}, 
      author={Tianhao Fu and Bingxuan Yang and Juncheng Guo and Shrena Sribalan and Yucheng Chen},
      year={2026},
      eprint={2603.13027},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2603.13027}, 
}

Dataset Download

You can download the dataset from Project Neura's Central Data Server (CDS).

Alternatively, you can download the dataset using MIP Candy:

from mipcandy import download_dataset

download_dataset("atatc/ut/esc102/SortScrews", "directory/to/save/dataset")

Codebase Installation

Our codebase contains necessary utilities to use the dataset.

pip install git+https://github.com/ATATC/SortScrews

To customize your own dataset, train your own model, or run inference using our codebase, please clone the whole repository.

Class Indices

We rank the screws by their lengths. Class indices are assigned from 1 to 6, from left to right.

Class 0 is reserved for the background.

Customization

You could use "collect.py" to collect your own dataset.

Key	Usage
C	Capture current frame
X	Remove last entry
0 - 9	Set class ID
Q	Quit

To append an existing dataset, simply replace

app = Collector()

with

app = Collector(append_from=LENGTH_OF_EXISTING_DATASET)

Acknowledgement

We kindly ask other Praxis II groups not to copy our solution. We do not permit other groups to use our dataset.