PyVision

Ready-to-use implementations of some of the most common computer vision algorithms.

In PyTorch only!

Currently available architectures

Multi Task Cascaded Convolutional Neural Network (MTCNN) : A SOTA face and facial-landmark detection architecture. Check out this out for more details.
YOLOv3: The SOTA object detection algorithm. For more details, read the docs.
FaceNet: A Unified Embedding for Face Recognition and Clustering: One of the most popular architectures used for facial recognition. For more details, check here.
DEtection TRansformer (DETR): An end-to-end object detection architecture using transformers. For more details, check here.

For full list of architectures that has been ported or are in the process of being ported, check here.

Run the code in your terminal to clone the master branch which contains the working code

$ git clone https://github.com/pranjaldatta/PyVision.git --single-branch --branch master

Then, go to the repository root by pasting the command given below into your terminal

$ cd PyVision

Run the following command in the terminal to install PyVision into the current virtual or conda environment

$ pip install .

For contribution guidelines, please look here. Contributions are always welcome!

Populate with more architectures (obviously)
~~Come up with an efficient way to make the repository minimal i.e. assets (like weights) will only be downloaded on as-you-need basis.~~ All weights are hosted on SRM-MIC Google drive and downloaded using gdown
~~Come up with an efficient way to ensure that heavy architecture specific dependecies are installed only when required.~~ All heavy assets are installed only when model is being used.

Currently, its working only in pre-configured conda environment with all dependencies installed.