Development of an AI that restores the “image you see” from brain activity
In recent years, AI-based image generation technology has made remarkable progress.
recently,A technology that records brain activity while viewing a still image, converts the visual information into AI words, and reconstructs the image.was reported.
For example, just by looking at a picture of a cat, AI can guess what the picture is based on brain activity alone, without knowing what we are looking at.
at the moment alreadyImage restoration with a high accuracy of 80%has been successful.
This news was also reported on our website.
Developed an AI that scans the brain activity of people and visualizes what they see!
On the other hand, only a small number of studies have been conducted to reconstruct the “moving images” that we watch from brain activity data.
The research team tried to develop this dream technology this time.
In the experiment, while having the subject watch the video sample“Magnetic Resonance Functional Imaging (fMRI)”is used to record the movement of brain activity in real time.
However, since fMRI captures snapshots of brain activity at intervals of several seconds, imaging only this will result in poor quality images.
This is because “30 frames per second” is generally the standard for images that look smooth.
Therefore, the team newlyAI model that learns spatio-temporal information from fMRI dataMind-Video」developed.
With conventional technology, brain activity data was converted into static images as they were.Mind-Video learns the spatio-temporal information of fMRI data to fill in the gaps between intermittently obtained images and connect them smoothly.can do.
And as a result of continuing training while having the subject watch the video sample,Mind-Video can now recover the visual information and motion dynamics of a video with a high accuracy of 85%.It is.
Here is a video sample (left) viewed by the subject and a video generated by Mind-Video (right).
The researchers say, “We were able to reconstruct basic objects, animals, humans, and scene situations with high accuracy from brain activity scan data alone.”
He continued, “More importantly, we were able to correctly reconstruct the dynamics of the video, including movements such as running, dancing, and singing, fast-moving scenes, and long shots of the cityscape.”
It is amazing that such accurate images can be reconstructed using only brain activity data.
Although it is not possible to output exactly the same image, the type of animal being watched, the location, etc. match, and when looking at the crowd, the scenery of the crowd is reproduced properly.
Is it possible to save the “dream” seen during sleep as a video?
Researchers believe that if this technology is further developed and put into practical use,It may be possible to record images of the thoughts you have in your daily life, or the dreams you had while you were asleep.said.
On average, we dream up to six times a night, but 90% of them are forgotten within minutes of waking up.
Dreams are an important place to catch creative ideas.
For example, the melody of the Beatles’ classic “Yesterday” is said to have come to Paul McCartney in a dream.
There are many other artists who have used the content of their dreams in their works, but there is no point in forgetting the dream you had.
But what if we could record brain activity while we were asleep and show it to AI the next morning?
You may find it easier to come up with novel ideas that you would never have thought of when you were awake.
#Developed #create #videos #scenes #brain #activityVisualization #dreams #future #Nazology