COIL-100 (Columbia Object Image Library)

Overview

COIL-100 contains 7,200 images of 100 physical objects. Each object has 72 RGB views, photographed at 5-degree increments around a full rotation. The distributed images are 128×128 pixels.

The only available split is train (7,200 examples). It represents the entire official dataset: Columbia does not provide a predefined train/test classification split.

Data Structure

Each example is a dictionary with these keys:

Key	Type	Description
`image`	`PIL.Image.Image`	128×128 RGB image
`label`	int	Zero-based object identity: original object 1 maps to 0, and object 100 maps to 99. Class names are `obj1` through `obj100`.
`angle`	int	Viewpoint in degrees, from 0 to 355 in steps of 5

Usage Example

from stable_datasets.images import COIL100

ds = COIL100(split="train")
print(len(ds))  # 7200
sample = ds[0]
print(sample.keys())  # dict_keys(['image', 'label', 'angle'])
print(sample["label"], sample["angle"])  # 0 0

References

Official Columbia COIL-100 page: https://www.cs.columbia.edu/CAVE/software/softlib/coil-100.php
Official archive: https://www.cs.columbia.edu/CAVE/databases/SLAM_coil-20_coil-100/coil-100/coil-100.zip

Columbia does not state a clear SPDX-style license on the download page. Check with the data provider for applicable use terms.

Citation

@techreport{nene1996coil100,
  author = {Nene, Sameer A. and Nayar, Shree K. and Murase, Hiroshi},
  title = {Columbia Object Image Library (COIL-100)},
  institution = {Columbia University},
  number = {CUCS-006-96},
  month = feb,
  year = {1996}
}