To download, please save the links below. This server supports resumes on downloads.

IMPORTANT: Browsers sometimes have problems to download the collection in one large file without corrupting it.
A download manager (Firefox addon Downthemall is known to work with large downloads) may help

MIRFLICKR-25000 (released in 2008 aka mirflickr08 or mirflickr25k)

1. Image collection (incl. tags, EXIF), approx 2.9 Gb

Collection in 1 large file (md5: A23D0A8564EE84CDA5622A6C2F947785)

2. Annotations


Please refer to README.txt for details.


One million images. The first 25000 images in this collection correspond to the MIRFLICKR-25k set

1. Download the 1 million images

Suggestion: Use Firefox and the DownThemAll! extension/addon for safe downloading

Directory containing 1 million images in ten 12GB files (images0, images1, ..., images9)

1.1 BitTorrent file for the 1 million images
with 10 tar files, each containing 10 directories of 10000 images; recommended BitTorrent client: uTorrent

2. Thumbnail images

3 zip files containing 100 directories with 10000 thumbnail images each
Directories 1 - 33: thumbs1-33.zip
Directories 34 - 66: thumbs34-66.zip
Directories 67 - 100: thumbs67-100.zip

3. Metadata - Tags, EXIF, license

Tags: tags.zip
Raw tags: tags_raw.zip
EXIF data: exif.zip
Raw EXIF data: exif_raw.zip
Licenses (incl. owner data, image titles, URLs): license.zip

4. Metadata - Visual Descriptors

100 files with descriptors for 1000 images each. More detailed documentation will be added shortly.
MPEG-7 Edge Histogram descriptors: Edge Histogram Descriptors
MPEG-7 Homogeneous Texture descriptors: Homogeneous Texture Descriptors.zip

5. Software - Visual Descriptors and Classification

TOPSURF: open source, easy to use API for bag of words similarity and classification
Visual Studio | Eclipse

RetrievalLab: a matlab-like tool for analyzing and exploring visual information retrieval
Windows Executable

Older versions