New H-FraDS method optimizes Vision Transformers for edge GPUs in autonomous vehicles

Researchers have developed a new hardware-aware scheduling methodology called Heterogeneous Frame Dispatch Scheduling (H-FraDS) to optimize the deployment of Vision Transformers on heterogeneous edge GPUs for autonomous vehicles. This approach addresses limitations in hardware utilization and accelerator-incompatible operators by routing frames across GPU and DLA cores. The adapted model maintains a high F1 score, and the H-FraDS Balanced Dispatch configuration achieved a significant speedup, meeting real-time operation requirements. AI

IMPACT This research could lead to more efficient AI processing in autonomous vehicles, enabling real-time performance with lower power consumption.

RANK_REASON Academic paper detailing a new methodology for optimizing AI model deployment on edge hardware. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Ashiyana Abdul Majeed
• Heterogeneous Frame Dispatch Scheduling
• NVIDIA
• Swin Transformer
• Vision Transformers

• infra
• paper

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →