[not issue, only questions] Is this tech "one neural nets for all" or one neural nets for capita? about mind-vis HOT 3 OPEN

I think you're being overly pessimistic @tobyxdd

Reading their documentation and trying the code out on publicly available datasets the code seems to work much better than previous solutions.

they even specifically mention that they tried to solve overfitting problems and tested on multiple datasets for validation.

The training data is diverse enough to be a good generalization for general- human vision.

However cross-person training still not being possible is a real problem , I would still say full training isn't needed for this to work on another person , just fine-tuning that would take less than 5% of the initial training time , judging by how this is built up.

from mind-vis.

I'm afraid this is not as advanced as you might think.

First of all this only reads what a person's eyes are currently seeing, not what they are "picturing" in mind. Secondly, these models rely heavily on pre-trained images and I seriously doubt they can handle any content significantly different from the training set. And lastly, yes, you must re-train the model for each individual.

from mind-vis.

And lastly, yes, you must re-train the model for each individual.

Thanks. I asked this detail because I'm curious what still would be needed to be done after the surgery to have a neuralink or anything similar. It seems no matter what the surgery can give me, I still have to do a lot per individual job to get it to work for "me", which may limit the usage of such inplants.

from mind-vis.