COIL-20 is a classic computer vision dataset containing grayscale images of 20 distinct physical objects. The dataset is widely used for object recognition and classification tasks. It was created by the Columbia University Image Library.
Use Cases
- Classify object_id from grayscale image features using convolutional neural networks.
- Benchmark feature extraction algorithms on images of 20 distinct objects.
- Train models for pose estimation using images of objects captured from multiple angles.
- Evaluate dimensionality reduction techniques on high-dimensional image data.
Strengths
- Standardized benchmark dataset with 20 distinct object classes.
- Images captured under controlled conditions with systematic pose variation.
Limitations
- Dataset contains only 20 objects, limiting model generalization to broader categories.
- Images are grayscale, lacking color information present in real-world scenes.
Provenance
- Source
- Columbia University Image Library (COIL)
- Collection Method
- Images captured of physical objects on a rotating turntable.
- Time Range
- null
- Freshness
- null
- Geography
- null