Our AI-Powered Photo Search is based on the research of OpenAI, and uses GPT-like natural language processing combined with state-of-the-art visual understanding to provide a superior photo search experience.
Our AI-powered photo search is fully trained to analyze your photos and find the ones that match your search term. We use pre-trained models that have been optimized on a large data set to accurately recognize and categorize different types of visual content.
Here's how it works:
We use pre-trained AI models: We utilize AI models that have already been trained on large external datasets of images and texts. We do not further train the models based on your images, ensuring a completely secure service that also takes privacy into special consideration.
We convert images to vectors: When you upload images to Jottacloud, our AI model automatically analyzes the content of each image and creates a vector representation of each image. A vector is a matrix of hundreds of numbers that cannot be read by humans, but which makes it possible to compare one vector to another.
We convert the search query to a vector: When you describe the image you are looking for, we also convert the search term into a vector representation.
We compare the vectors: Now we calculate the distance between the search vector created from your search and the image vectors associated with each image you have stored with us. If the distance between the vectors is small, it means that the vectors are similar. Then we say that the similarity score between the two vectors (the search and the image) is high.
We show you the result: We sort the images from highest to lowest similarity score, so you can see the most accurate and relevant images based on your search request.
We do it quickly: From the moment you search until the result appears, it takes 0.3 seconds. Pretty cool, right?
To learn even more about the technology behind our AI-powered Photo Search, read our blog post.
Current limitations of our AI Powered Photo Search:
It currently doesn't have personalized facial recognition, i.e. the option to tag specific people with names. The model does, however, understand and recognize faces, and by using the "Find similar" tool, you can find images that have the same people in them. Enabling people tagging is something we will most likely add in the near future!
The location of the picture which can be found in the metadata is yet not integrated into the search, which means geographical searching is limited. However, the model recognizes famous locations, cities and regions, so searching for "The Eiffel Tower in France" or "Taj Mahal" will give very accurate results. It can also understand whether pictures are taken in New York or London, as the model has been trained on millions of images.
It currently doesn't support searching for time periods
It is limited by the visual content of your photos, which may not always accurately represent what you search for (if the quality of the image is poor, the angle is strange, etc.)