CUDA Object Detection

Authors

• Alexandre Lemonnier
• Victor Simonin
• Adrien Barens
• Sarah Gutierez

---

Usage

Build

For building the project, you need to have CMake installed. In cpu_implem_opencv, cpu_implem and gpu_implem directories, run the following commands:

cmake -B build
cd build
make

Then, for running the program in cpu_implem and gpu_implem, you can use the following command:

./program_name [--save] <image_ref_path> <image_test_path> [image_test_path2] [image_test_path3] ...

Generate frames

To generate frames from a video, you can use the script in the tools directory:

python3 frame.py <video_name> <video_path>

Visualize

For visualizing the bounding boxes from the output json, you can use the python script visualize.py in the tools directory.

cat output.json | python visualize.py

You can choose to run one of the implementation in the object_detection.py python script on an image folder.

python object_detection.py <image_folder_path>

Benchmark

To generate the benchmark, go into the benchmarks directory and run the following command:

cmake -B build
cd build
make
./benchmark

or with

python python/benchmark.py