Full record view help

Similar images

An image similarity search was integrated using image vectors that were created on the basis of the self-supervised learning algorithm SwAV (Swapping Assignments between Views). The algorithm is described in Caron et al. [1], an implementation is provided by Facebook [2]. A SwAV model [3] pre-trained with the ImageNet data set [4] was used as a model. Resulting image vectors, which are shortened to 80 dimensions sufficient for the result, were precalculated for all images in the image archive and saved in the index. The search engine queries are reduced to calculating the distance between the vectors stored in the index. The smallest distance is calculated using the Euclidean distance.


  1. Mathilde Caron and Ishan Misra and Julien Mairal and Priya Goyal and Piotr Bojanowski and Armand Joulin (2020). Unsupervised Learning of Visual Features by Contrasting Cluster Assignments. CoRR, abs/2006.09882.
  2. https://github.com/facebookresearch/swav
  3. https://pl-bolts-weights.s3.us-east-2.amazonaws.com/swav/swavimagenet/swavimagenet.pth.tar
  4. https://www.image-net.org/