Alveo U55C Deployment: ResNet18/50 (CIFAR-100) and ResNet20 (CIFAR-10) with Vitis-AI & HLS(hls4ml) Comparison

The project deploys ResNet architectures on the Xilinx Alveo U55C using Vitis-AI, and compares the performance with a custom HLS accelerator using hls4ml.

• ⚡ Vitis-AI Deployment:

  • Implementation of ResNet18 and ResNet50 on the CIFAR-100 dataset.
  • Implementation of ResNet20 on the CIFAR-10 dataset.
  • Utilizes the Vitis-AI DPU (Deep Learning Processor Unit) overlay for high-throughput inference.

• 🛠️ HLS Custom Implementation:

  • Implementation of a custom hardware accelerator of ResNet20 using hls4ml.
  • Contrasts the DPU accelerator against a custom FPGA accelerator.
  • Source codes and synthesis scripts for this custom implementation are located in the hls4ml/ folder.