Machine Vision Algos

The machine vision algorithms repository contains implementations for the following algorithms:

• Canny-End-To-End: Edge detector

Setup

cd <project directory>
conda env create --file python/conda_env.yml --prefix .conda

Canny-End-To-End

See more python/canny.

python/canny/canny_playground.ipynb is an interactive notebook that allows one to play around with the canny algorithm.

python/canny/canny_impls contains multiple implementations that can be selected in the notebook.

Numba Cuda FP32

This implementation runs on cuda using numba. It runs on nvidia gpus.

The playground notebook has toggle at the top to enable a cuda simulator. The simulator runs very slowly but works!

Find it here rd_numba_cuda_fp32.py.

Vec v4 dibit

This is a vectorized canny implementation. It uses only numpy and scipy. It runs on the cpu.

Find it here rd_vec_v4_dibit.py.

TODO Implementations

• enable multiprocessing for opencv and numba on cpus
• numba stencil
• numba normal jit with loops
• opencv2 only

TODO

• Benchmark notebook: Add system information like gpu, cpu, ram, ...
• Visualize gauss radius outline for sigma sliders

Image stitching and points of interest

See more python/points_of_interest.

python/points_of_interest/points_of_interest.ipynb is an interactive notebook that allows one to play around with image stitching and points of interest algorithms.

It implements the following algorithms:

• Harris corner detector
• Descriptor computation
• Random sample consensus (RANSAC)
• Image stitching using SIFT and FLANN

Contributions

Contributions are welcome! Open an issue before working on something big or complex.

Usage as teaching material

Feel free to use anything in this repository as teaching materials. If you do, I would love to be mentioned as a source. An email informing me would also be appreciated.

Thank you.