CORnet v1.1 wrapped for Brain-Score testing.
This is a fully recurrent version of CORnet-S, meaning that each layer is now recurrent such that data is processed at each layer at each time step. In contrast, in CORnet-S data is processed at a single layer at each time step; once the data is processed for several time steps in a given layer, it is passed to the next layer, while the previous layer sits idle.
| Benchmark | Score | CORnet-S score |
|---|---|---|
| FreemanZiemba2013.V1-pls | 0.249 | .294 |
| FreemanZiemba2013.V2-pls | 0.304 | .242 |
| MajajHong2015.V4-pls | 0.567 | .58 |
| MajajHong2015.IT-pls | 0.511 | .54 |
| Rajalingham2018-i2n | 0.338 | .55 |
| Kar2019-ost | N/A | .305 |
Notes:
- The model has not been prepared for testing on
Kar2019-ostso the score is unknown. - The score on
Rajalingham2018-i2nis suspiciously low. I'm pretty sure it's incorrect. - Other scores match my expectations. The model was not developed sufficiently to reach the level of CORnet-S but should be very close.
I am not actively developing this model and have no plans to do so. Feel free to use this model for your needs.