We’re thrilled to introduce our new major release today: pixolution flow 4. This is a complete rewrite of our visual search engine, with massive improvements and radical innovations.
In this article you’ll learn how all core functionalities have been improved in their performance, how our technology can adapt even better to individual needs thanks to the possibility of custom AI modules, and we’re going to introduce a new feature that opens up new use cases.
pixolution flow 4 is based on an improved AI model we built. We used a deeper model and trained it with ten times more concepts than our previous version.
With this deeper insight into the content of images we are able to noticable improve the visual search experience. Compared to our previous product version we can now provide more relevant results for visual search queries. The precision of duplicate detection is also significantly improved. It can cope with more variations and is more robust against image modifications.
Our image tagging approach is based on the retrieval of the most similar images that are already tagged and part of your collection. Improvements of our visual search therefore have a direct impact on the quality of tagging predictions. This way pixolution flow 4 helps you to tag new images you want to add to your image collection. And this is done by using your language and specific wording style of your metadata.
If you have a specialized image collection you would like to classify or adapt visual search to subtle differences in images, we can help you even better now.
This is possible because pixolution flow 4 is based on a completely new modular architecture. pixolution flow is a hub for modules adding features. For example, the visual search, the multi-color search, or the text space filter are encapsulated in modules.
Modularity is the key. We are now able to build and integrate custom AI modules into pixolution flow 4. In short, it works like this: You tell us your needs and we train an individual AI model with your data. The resulting module can be easily added to your pixolution flow instance providing the custom functionality like specialized analysis and search features.
2. Using Artificial Intelligence to detect COVID-19
3. Real vs Fake Tweet Detection using a BERT Transformer Model in few lines of code
4. Machine Learning System Design
The possibilities here are endless. For example, we could integrate a module for recognizing certain objects and logos, or to detect persons in images, or a module that classifies image content based on specific categories.
Drop us a line and we’ll talk about your individual wishes for a custom module.
In pixolution flow 3 we extended the text search engine Apache Solr with visual search capabilities, but the visual search was limited to only one image representing one document. pixolution flow 4 now supports multiple images that visually represent one document.
This opens up some awesome new application possibilities. For example, you can index a series of images or different versions of images to make them searchable. Long videos with changing content can also be fully indexed and searchable.
These application scenarios are possible now:
When an image contains more than just one object it is often not clear what is actually the important part in the image. For example it can be as small as a ring.
With pixolution flow 4 you can handle this and index close-up images as well. If a user in an online shop searches for a specific ring, pixolution flow 4 will find one of its close-ups and you can display the associated main image. This is a fantastic user experience.
Search for Products and 3D Objects
If you have product shots from different angles, you can index all product shots to achieve a more robust search quality when it comes to 3D objects like shoes or cars.
Instead of indexing just one master keyframe per video, you can now index several frames. This way videos with changing scenes and changing content can be made fully searchable.