Region Layer Commitment
BaseModel layers have to be committed to cortical regions.
For BaseModels that are automatically translated into BrainModels
this is done on separate public data.
The same layers are thus used when recording from the same cortical region,
e.g. always the same layer for V1 instead of different layers per benchmark.
By default, committing layers to regions is done with a layer selection on separate public data:
each model is tested on its alignment to brain data from each region (V1, V2, V4, IT)
and the best layer is committed to that region.
At the end of this step, the model thus has a fixed, unchangeable layer commitment
for each region.
Visual Angle
Models have to declare their field-of-view so that stimuli can be displayed like they were
displayed to experimental subjects.
For instance, if experimental stimuli were shown at 4 degrees and a model's field-of-view is
larger than that, then the stimuli are padded such that the core stimulus will make up 4 degrees
in the model's field-of-view.
Here's an example for a model declaring 8 visual degrees as its input.
Base image, presented at 8 degrees:
If image is presented at 4 degrees, model sees:
If image is presented at 12 degrees, model sees:
Stuck?
Our tutorials and FAQs, created with Brain-Score users, aim to cover all bases. However, if issues arise, reach out to our community or consult the troubleshooting guide below for common errors and solutions.