| Autor: |
Ozimek, Piotr, Siebert, J. Paul |
| Jazyk: |
angličtina |
| Rok vydání: |
2017 |
| Předmět: |
|
| Zdroj: |
BMVC 2017 Workshop on Deep Learning on Irregular Domains |
| Popis: |
We present a biologically inspired method for pre-processing images applied to CNNs\ud that reduces their memory requirements while increasing their invariance to scale and rotation\ud changes. Our method is based on the mammalian retino-cortical transform: a\ud mapping between a pseudo-randomly tessellated retina model (used to sample an input\ud image) and a CNN. The aim of this first pilot study is to demonstrate a functional retinaintegrated\ud CNN implementation and this produced the following results: a network using\ud the full retino-cortical transform yielded an F1 score of 0.80 on a test set during a 4-way\ud classification task, while an identical network not using the proposed method yielded an\ud F1 score of 0.86 on the same task. The method reduced the visual data by e×7, the input\ud data to the CNN by 40% and the number of CNN training epochs by 64%. These results\ud demonstrate the viability of our method and hint at the potential of exploiting functional\ud traits of natural vision systems in CNNs. |
| Databáze: |
OpenAIRE |
| Externí odkaz: |
|
