NTUT EE dept, CE team Master Student
An AIoT project based on PYNQ-Z2 FPGA Evaluation board. Reading image from usb camera and running yolov3-tiny detection with DPU and using MJPEG HTTP Streaming. Also provide an Android app for handy accessing the streaming content.
Optimize the yolov3 model to speed up inference time by 83.9% for deploying on edge devices.
This project is still in progress. This is the phase 1 progress report. slide
2021-09-27 - Deploying the yolov4 detector with Vitis-ai v1.4, demo with video: link
Sparse training - Finished layer and channel pruning - Finished Finetuning pruned model. - Finished Quantizing model with nndct