Smart imagers modeling and optimization framework for embedded AI applications
http://api.archives-ouvertes.fr/search/?fq=halId_s:cea-04548812&fl=title_s,en_title_s,docid,label_s,en_label_s,authIdHal_s,halId_s,structId_i,uri_s,keyword_s,en_keyword_s,authLastNameFirstName_s,journalTitle_s,abstract_s,en_abstract_s,producedDate_tdate,producedDateY_i,language_s&sort=publicationDate_s+desc
authors
keywords
- Smart imagers
- Behavioral simulation
- Spiking imagers
- Embedded artificial intelligence
abstract
This work presents a framework for behavioral simulations of smart imagers with hardware and power constraints. The objective is to compare innovative imaging systems that would be composed of a specific image sensor and a dedicated image processing. For that purpose, a versatile imager model is presented and applied to a time-to-first-spike imager associated with two types of neural networks. Image classification is targeted to assess the system performance, namely the classification accuracy and data throughput. Simulation results depict/show the impact of different key-parameters helping in the choice of the final imaging system architecture.
Aix Marseille Université | All rights reserved - Legal notice