Image classification and object detection datasets containing 11,788 images consisting of 200 species of birds (Warning: Images in this dataset overlap with images in ImageNet).

• Source (README)

• Original dataset (1.2GB)

• Image classification dataset (1.1GB)

• Object detection dataset (VOC XML format) (1.1GB)

• Object detection dataset (ADAMS format) (1.1GB)

Conversion from original data:

• Rename images folder to subdir. This is the image-classification form of the dataset.

• Run the attached Python script (requires the wai.annotations library; tested with 0.7.5) in the CUB-200-2011 directory to parse the bounding_boxes.txt into Pascal VOC format. The data in the voc directory is the object-detection form of the dataset.

License (according to Caltech website)

We do not own the copyrights to these images. Their use is restricted to non-commercial research and educational purposes.