This work presents CascadeCNN, an automated toolflow that pushes the quantisation limits of any given CNN model, aiming to perform high-throughput inference. A two-stage architecture tailored for any given CNN-FPGA pair is generated, consisting of a low- and high-precision unit in a cascade. A confidence evaluation unit is employed to identify misclassified cases from the excessively low-precision unit and forward them to the high-precision unit for re-processing. Experiments demonstrate that the proposed toolflow can achieve a performance boost up to 55% for VGG-16 and 48% for AlexNet over the baseline design for the same resource budget and accuracy, without the need of retraining the model or accessing the training data.
@article{arxiv.1807.05053,
title = {CascadeCNN: Pushing the Performance Limits of Quantisation in Convolutional Neural Networks},
author = {Alexandros Kouris and Stylianos I. Venieris and Christos-Savvas Bouganis},
journal= {arXiv preprint arXiv:1807.05053},
year = {2018}
}
Comments
Accepted for publication at the 28th International Conference on Field Programmable Logic & Applications (FPL), 2018