How do I go from a collection of images, to a full machine learning dataset, that I can then use for transfer learning (resnet50 etc.)?

Question

I'm new to the field of machine learning and just starting to get the hang of it. I was wondering how I can go from an amount of images, to a machine learning dataset. I'm not quite sure on how to store the labels for my data either? A csv format?

Answer 1

This is a rough description on how the full workflow would look like. I can't give much more detail since your question is not very specific.

1. Label your images

There are many tools to label a set of images. You need to find one that fits the deep learning framework you want to use to train your neural network.

I've used OpenLabeling to label my dataset in the YOLO format. However, this might not be the format you're looking for.

2. Find a pre-trained model

If you want to do transfer learning, you need a model (neural network) that has already been trained on another dataset.

3. Setup your PC

You need to install everything you need for training the neural network. I also recommend you to install and use CUDA (use your GPU to train the network instead of your CPU) since it's a lot faster.

4. Train your model

Use the transfer learning functinallity of your chosen framework to train your network using the pre-trained model and your labeled images.