High-Performance Deployment of Text Detection Model: Compression and Hardware Platform considerations
2022 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS)(2022)
摘要
Network compression is often adopted for high throughput implementation on commercial accelerators. We propose a heuristic based approach to obtain compressed networks with a hardware-friendly architecture as an alternative to conventional NAS algorithms that are computationally expensive. The proposed compressed network introduces 142 $\times$ memory-footprint reduction and provide throughput improvement of 5-8 $\times$ on target hardware platforms, while retaining accuracy within 5% of the baseline trained model. We report performance acceleration on CPU, GPU, and FPGAs for a text detection task.
更多查看译文
关键词
Lottery-Ticket Hypothesis,Knowledge Distillation,Performance Improvement
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要