Internet offers many possibilities to gather specific images to train models. For example, Google image searches allow to collect various images for one subject. PyImageSearch blog provides excellent guides on building a deep learning dataset using search engines:

Manually scraping Google Images: https://www.pyimagesearch.com/2017/12/04/how-to-create-a-deep-learning-dataset-using-google-images/ Automatically Using Microsoft’s Bing Image Search API: https://www.pyimagesearch.com/2018/04/09/how-to-quickly-build-a-deep-learning-image-dataset/