one option to use semi-cooked features from image data

Dear Saif ul Islam

I would Suggest to look for Prototypical networks, matching learning and etc, for few shot learning problems.

I recently studied bellow paper which was good:

https://papers.nips.cc/paper/2017/hash/cb8da6767461f2812ae4290eac7cbc42-Abstract.html

Dear Saif ul Islam

These articles might be helpful, have a look:

1. https://www.tno.nl/en/focus-areas/artificial-intelligence/application-areas/challenges/learning-from-small-datasets/

2. https://www.kaggle.com/rafjaa/dealing-with-very-small-datasets

3. https://towardsdatascience.com/how-to-use-deep-learning-even-with-small-data-e7f34b673987

Kind Regards

Qamar Ul Islam

Dear Saif ul Islam

If your looking for a CNN for grayscale images (small dataset), check this article:

https://www.kaggle.com/sujoykg/keras-cnn-with-grayscale-images

Transfer learning is a good way to overcome it. you may use initial weights from already trained algorithms on a little different dataset and use those weights to start your training.

One more example but on classification task.

Conference Paper Comparison of Combinations of Data Augmentation Methods and ...