Meta-Album Cars: 16,185 Images of 196 Car Models for Fine-Grained Classification

Meta_Album_CRS_Extended is a preprocessed version of the Stanford Cars Dataset, containing 16,185 square images across 196 distinct car model classes. The dataset was originally collected in 2013 by researchers including Jonathan Krause and Li Fei-Fei. Images were resized to 128x128 pixels using an anti-aliasing filter after being squared by duplicating border pixels to preserve information.

Use Cases

• Benchmarking few-shot image classification models based on the 196 fine-grained car model classes.
• Training models for fine-grained visual categorization of vehicles using the preprocessed 128x128 pixel images.
• Developing data augmentation or preprocessing techniques for image datasets where preserving object information during resizing is critical.

Strengths

• Contains 16,185 images, providing a substantial volume for model training.
• Offers 196 fine-grained classes, enabling detailed categorization tasks.
• Images are consistently preprocessed to a uniform 128x128 pixel resolution.

Limitations

• The original data is from 2013, which may not reflect newer car models.
• Description metadata is limited; actual data quality requires manual inspection after download.
• The preprocessing method of duplicating border pixels may introduce artifacts not present in the original images.

Provenance

Source

Stanford Cars Dataset (https://ai.stanford.edu/~jkrause/cars/car_dataset.html)

Collection Method

Images were collected, then preprocessed by creating square images via border duplication and resizing to 128x128 pixels.

Time Range

Original data collected in 2013.