Examples: Update documentation and support files for examples to account for new deployment documentation

[hbooth]

Definition of Done

• [ ] Documentation landing page for examples
• [ ] Documentation updated to point to deployment documentation
• [x] Spurious files removed

READMEs answer the question of "What do I do next?" Documentation of example will walk a user through the particulars of an example. Tutorial of the example. Documentation page for examples will describe how to accomplish a particular task or set of tasks or describes the nature of examples/plugins/etc..

Examples Landing Page

Prerequisites Setup

1. point to deployment instructions
   • (mention cpu/gpu distinction)
2. point to dataset download instructions
3. Instructions for mounting datasets for a local/single machine installation
   • point to more general instruction.
4. Instructions for minimal python env/dependencies (point to script)
5. Usage Instructions
   • (port make file to python script)

Table Summarizing Example

Dataset used (prereq) GPU needed? (prereq) What does it show? Attacks/ Defenses (info) ML Librrary? pytorch/tensorflow (info) architecture (info)

Summary of datasets and structure

Each of the examples is updated:

• [x] pytorch-detectron2-demo
• [x] pytorch-mnist-membership-inference
• [x] tensorflow-adversarial-patches
• [x] tensorflow-backdoor-poisoning
• [x] tensorflow-imagenet-pixel-threshold
• [x] tensorflow-imagenet-resnet50-fgm
• [x] tensorflow-mnist-classifier
• [x] tensorflow-mnist-feature-squeezing
• [x] tensorflow-mnist-model-inversion
• [x] tensorflow-mnist-pixel-threshold

These examples are marked for deletion

• [x] basic-mlflow-demo
• [x] patch-defended-pixel-threshold
• [x] pytorch-robust-dpatch
• [x] tensorflow-roadsigns-yolo

Tasks for each example

• Task plugins go into common folder under examples directory
• Script to register all task plugins
• rename to avoid collisions and indicate example
• move MLproject files to src and remove use of relative path to src
• Delete extraneous files
  • init.db
  • environment.yml
  • architecture diagram
  • utils.py
  • /data subdir
  • /jupyter subdir
• simplify notebooks to focus only on example
• remove any descriptions of dataset

[jkglasbrenner]

Originally posted by @jkglasbrenner in https://github.com/usnistgov/dioptra/issues/19#issuecomment-1462671488

Figured it out, the 403 - Forbidden error appears if you don't agree to the competition rules (even though the competition is over). At the bottom of https://www.kaggle.com/competitions/imagenet-object-localization-challenge/data, you'll see this:

People need to sign-in using the account they used to generate the Kaggle API token, and then the "Sign In" will be replaced with I Accept/I Agree. Once you agree, that box gets replaced with:

Now if you try to download using the kaggle CLI tool, it will work (provided you remember to generate the API token).

[jkglasbrenner]

All examples have been refactored and merged into dev, closing.