About this document

FPGA-Aware Acceleration for Vision Transformers by chaitanyabphc03 is a document available to read on EtoBox.

The document presents Auto-ViT-Acc, an FPGA-aware automatic acceleration framework designed for Vision Transformers (ViTs) utilizing mixed-scheme quantization to enhance performance while maintaining accuracy. The framework combines fixed-point and power-of-two quantization methods, achieving significant improvements in frame rates on FPGA hardware with minimal accuracy loss. This work is notable for being the first to explore model quantization for ViTs in an FPGA context, offering a comprehensive solution

Author

chaitanyabphc03

Language

EN