Research tools

[nbourne2]

Use OpenCV library.

[nbourne2]

Tools to consider:

• pyTorch (used previously for GAN development in super-modis)
• Keras/Tensorflow (used previously for SR in super-sen2 and LSTM in water-watch)
• OpenCV, an open source library of over 2500 algorithms
• DeepForest, a python package for airborne object detection and classification, by Weinstein et al. 2019

While I have built models previously using frameworks like pyTorch and Keras, it looks likely that higher-level frameworks more specialised for the task are now available in the form of OpenCV and DeepForest. OpenCV is widely used and well known, also probably has lots of tutorials on the website and on Medium etc, so would probably be a good one to get some experience with. DeepForest is more specialised and less well-known; another problem with using that is that it was used for generating the training data I'm looking at using in the first place.

Now I need to do some more research on what models and architectures are available through OpenCV, and how to use this library for training these models. Start with the OpenCV docs.

[nbourne2]

Another tool to explore is FiftyOne, an open-source tool for building high-quality datasets and computer vision models. This is useful for visualising and annotating training data, and for visualising and evaluating model predictions.