In this project, we design a system of obeject detection intending for self-driving car usage. The system is able to defense adversarial attack. The system can identify the inputs which are attacked and we use YOLO to verify the results. demo video
Considering some limitation of the development board, we demonstrate the system in two different scenario. The overview of each scenarios is shown below. For more information, please refer to the demo part of our demo video.
- Install Edge Impulse CLI
- Install ARC GNU ToolChain
- Donwlaod and Setup SDK
- Clone this repository to your local computer
- Modify the
ROOT_PATH
in scenario_1/Makefile and scenario_2/Makefile - Connect the WE-I to the computer by USB cable
$ cd scnario_1
- Place tflite file under
checkpopints/
directory. Pretrained model file is available at https://drive.google.com/file/d/1OZqE7vJ8KH-Pt2yfjFHCvaPEqgmCetMN/view?usp=sharing $ make
$ make flash
$ himax-flash-tool -f output_gnu.img
- Press the "reset" button on the board
$ python3 arc_detect.py #receive image from the baord and run YOLO inference
$ cd scenario_2
- Copy one of test_sample from test_samples/ to src/
$ make
$ make flash
$ himax-flash-tool -f output_gnu.img
screen /dev/ttyUSB0 115200 #see the result
- Verify the result using YOLO model with the images in test_sample_images/