GAN: Loading images, data augmentation and saving images

General
#1

Hi, I really like the structure of perceptilabs. I have a couple more questions:

  1. Is there any more info on how to load your own dataset of images? E.g What is the best way to save a folder of jpgs as a numpy array (.npy).

  2. Are there any tools for data augmentation? For small datasets it’s really nice to be able to rotate, flip and crop images to increase the training dataset.

  3. What is the best way to save images produced by the model?

Thanks!

#2

Hi @Sleach,

I’m glad you are liking the tool! :slight_smile:

  1. You can load data into the tool in a few different ways, you can see them all here: https://perceptilabs.com/docs/supported_data_files
    Loading a folder of jpgs works also, you can just select the folder and the Data component will do the rest.
  2. There are a few Processing components in the tool. Reshape also allows you to rotate the data and Rescale can be used to change the size of it. Crop and flip are incoming.
    Note that these components only process data going into the model and therefore can’t be used to increase the dataset.
    Something we are currently looking into is a “DataHub” where you would be able to do that kind of data augmentation and build data pipelines on your entire dataset. Before that is ready though, you unfortunetly need to do that kind of autmentation outside of PerceptiLabs or by using custom code in the Data Component.
    Also, is there anything more you would wish out of such a “DataHub”? :slight_smile:
  3. As PerceptiLabs currently is very focused on training a model, it does not poduce any images for you. We are looking at adding a small serving script with every export which would do that for you though. The best way right now would be to write up such a small serving script where it runs the model and saves the outputs in a desired format.
    You can also access any images from the training and test view by clicking on the download icon in the image.

Hope that answered your questions, and let me know if you have any more!
All the best,
Robert

1 Like