semihyagli/STL10-Segmented

STL10 - Segmentation

Please consider sponsoring this repo so that we can continue to develop high-quality datasets for the AI and ML research.

To become a sponsor:

GitHub Sponsors
Buy me a coffee

You can also sponsor us by downloading our free application, Etiqueta, to your devices:

Etiqueta on iOS or Apple Chip Macs
Etiqueta on Android

This repo contains segmented images for the labeled part of the STL-10 Dataset.
If you are looking for STL10-Labeled variant of the dataset, refer here: STL10-Labeled.
More information on the original STL-10 dataset can be found here.
Thanks to Martin Tutek, the original STL-10 dataset can be downloaded via the python code in this repo. For convenience, this code is copied in stl10.py in this repo.

If you use this dataset in your research please do not forget to cite:

@techreport{yagli2025etiqueta,
  author      = {Semih Yagli},
  title       = {Etiqueta: AI-Aided, Gamified Data Labeling to Label and Segment Data},
  year        = {2025},
  number      = {TR-2025-0001},
  address     = {NJ, USA},
  month       = Apr.,
  url         = {https://www.aidatalabel.com/technical_reports/aidatalabel_tr_2025_0001.pdf},
  institution = {AI Data Label},
}

@inproceedings{coates2011analysis,
  title     = {An analysis of single-layer networks in unsupervised feature learning},
  author    = {Coates, Adam and Ng, Andrew and Lee, Honglak},
  booktitle = {Proceedings of the fourteenth international conference on artificial intelligence and statistics},
  pages     = {215--223},
  year      = {2011},
  organization={JMLR Workshop and Conference Proceedings}
}

Note: If you notice any errors and/or if you have comments/ideas relevant to this dataset or Etiqueta in general, please reach me out at contact@aidatalabel.com.

Instructions

For Original Data:

You can download the stl10 image data by running

python stl10.py

This will:

1. create a folder named data download and extract the stl10 dataset inside that folder.
2. show one example picture in a new window. How to read images and their labels are also exemplified inside the stl10.py. For example, you can load all test images and their labels to a numpy array by using:

import stl10

img_test_X_bin_loc = "./data/stl10_binary/test_X.bin"
img_test_y_bin_loc = "./data/stl10_binary/test_y.bin"

test_X = stl10.read_all_images(img_test_X_bin_loc)
test_y = stl10.read_labels(img_test_y_bin_loc)

Additionally, other useful functions are readily defined inside stl10.py.

For Segmentation Data:

Segmentation of each image in the test_X.bin file can be found inside the provided .json files. Note that images that contain more than a single segment are segmented using different labels.
The combined segmentations can be recovered running:

python recoverSegmentations.py 

By default, this will create test_X_segmented.npy, which contains the cutouts of images in the test part of stl10 as depicted in the examples before.

• Depending on your device specifications, the above should take about ~10 minutes to complete. The segmentation progress will be printed out so that you can go get a coffee while the segmented data is being saved. Feel free to modify the script depending on your needs.

You can load this numpy array by using:

from numpy import load 

DEFAULT_SEGMENTED_TEST_X_SAVE_LOC = "./test_X_segmented.npy"

test_X_segmented = load(DEFAULT_SEGMENTED_TEST_X_SAVE_LOC)

Observe that the arrays inside test_X_segmented are now sparse.

Enjoy!

Examples

Class	airplane	bird	car	cat	deer
original
segmented

Class	dog	horse	monkey	ship	truck
original
segmented

Notes:

We have caught the following errors in the test part of the STL-10 dataset:

1495: cat_0 mark is in fact a dog_0.
6417: cat_0 mark is in fact a dog_0.
1718: cat_1 mark is in fact a dog_0.
1138: dog_1 mark is in fact a cat_0.
1484: dog_1, dog_2, and dog_3 are in fact sheep_0, sheep_1, sheep_2.
6566: dog_0 and dog_1 marks are in fact cat_0, and dog_0.
7902: dog_0 and dog_1 marks are in fact cat_0, and dog_0.