Build datasets for GAN, by scraping images, normalizing them & building classifier networks using Tensorflow or Pytorch

I want to create GANs for generating skulls in tribal style, with separate GANs for human skulls and animal skulls (i.e. generally ram/bull/cow skulls). First, I need to gather relatively large datasets for the purpose though, and clean up / normalize them (remove watermarks, background patterns and other irrelevant artifacts, crop them to a relevant size, etc). I need you to: 1. Build keyword-based image scraper code to gather a large number of samples 2. Manually filter out and categorize sets of each class large enough to use for start building the classifier 3. Build code for normalizing the samples, cropping etc 4. Build classifier code that can automatically find appropriate samples of each class & normalize them 5. Ensure that the final deliverable is able to automatically build a large dataset (at least 10 000 of each class), with a very small amount of false positives, that should be easy/painless to filter out manually. I have already scraped a large number of potential samples from a vector-based stock photo site, and can share that dataset as well as the code I wrote to perform the scraping (basically just a few lines of bash & perl), but want you to build and deliver your own scraping code to gather more samples from Google image search and other sources as well. Note that there are a large number of irrelevant images in the samples I've gathered, which is exactly the problem. I expect high quality documented code in your deliverables, no quick'n dirty copy/pasted code that I need to clean up myself. Initially, I just want datasets with white backgrounds and black lines for skulls, to avoid confusing the GAN I build. If you do this job well, I will most likely have similar projects for you in the future. When posting an offer, let me know about relevant experience in scraping images for datasets as well as building the code to normalize and classify the images, as well as which languages and frameworks you will be using.