Remember those 3D masks in the Mission Impossible movies? Using just a 2D image, they were able to create them from scratch effortlessly. Well, thanks to deep learning, we are not too far away from this fiction becoming a reality.
A UCLA research team has revealed a new technique that extends the capabilities of fluorescence microscopy, which will allow scientists to precisely label the parts of living cells and tissue with dyes that glow under special lighting. The researchers use AI to turn 2D images into stacks of virtual 3D slices showing activity inside organisms.
This research builds on an earlier technique Ozcan and his colleagues had developed that will allow them to render 2D fluorescence microscope images in super-resolution. Both these techniques advance microscopy by relying upon deep learning — using data to train a neural network, a computer system that is inspired by the human brain.
According to one of the developers, they are building a neural network classification model to identify the elements in an image that are the most dominant.
The initial resulting images range from impressive to strait funny. As you zoom into the image, you can see how pixels are broken down into different parts. It makes fascinating viewing. But as users keep uploading images, the model keeps learning with the increase in the training data.
Since this concept is still in its initial stages, the developers are working to continuously improve the understanding and accuracy of the model.
While the model itself and even the released tool are still in early development, one can imagine that this technology being applied in the gaming industry. Even video editors are currently using an amalgamation of tools, could find this to be an immensely helpful addition.
A few months, a similar AI was revealed back by researchers in the UK which converted 2D images of faces to 3D. But this tool does a lot more than just facial recognition. It looks at the entire image and then attempts to convert it into a 3D image.
That, as we mentioned, takes a massive amount of data to train and improve. It’s an audacious project being attempted but it opens the door for others to try it out as well.