Blog

Latest news and business insights

Release of pixolution flow 3.0.0

posted
  • 01 September 2016
  • Many months of hard work and developing and now it's finally there: Our major release pixolution flow 3.0.0. It ships with a lot of new features and optimizations. 

    New features

    Brand new image descriptor

    Image descriptor (formerly feature) is our new term to describe a pixolution representation of an image. We developed a new image descriptor which is very powerful being based on machine learning techniques. You will especially like that it greatly reduces the dependency on metadata.

    The new image descriptor is actually a container format for different smaller descriptors. Each of them describes different aspects of what’s important in an image. A whole image is represented in just 140 bytes regardless of image size. In comparison, a usual thumbnail has about 10,000 bytes. 

    New param rank.mode

    With the powerful new param rank.mode you can control the way images are scored and thus influence the result ranking. This easily enables you to optimize scoring based on your use case and image sets. pixolution flow ships with four presets: visual, balanced, semantic, and color. If these are not enough you can even define your own ranking mode.

    Smart Filter

    Searching in Solr is a two step process: collecting relevant images (number of found images) and ranking/scoring them. When doing visual similarity search the scoring process is expensive for millions of images. Our new smart filter drastically reduces the amount of images that have to be scored. The key is filtering out irrelevant images for dramatically faster response times. You decide how aggressive the filtering is done by setting the filter level (normal, high, very high, or extreme). The best thing about the smart filter is that is does not rely on any metadata.

    Powerful Command Line Interface

    Via the command line interface you can create image descriptors outside Solr. It's great that the command line interface now supports bulk processing and multithreading. This way it speeds up the process when calculating many image descriptors.

    Optimizations

    Better image search

    The new image descriptor of pixolution flow 3.0.0 excellently improves search quality. At the same time it reduces the dependency on metadata. Experience the new search quality in our online demo.

    Faster color search

    Color search is now faster since less data must be read from the index. We reduced the amount of bytes per image needed for color information by 65%, of course still ensuring consistent quality.

    Better and faster Keyword Suggester

    By default the Keyword Suggester now uses a smart filter and semantic rank mode. Look forward to faster results in huge collections and more accurate keyword suggestions.

    You would like to know more about pixolution flow? Get in touch with us!