Jonathan Prokos

See https://github.com/twosixlabs/armory/pull/1628#issuecomment-1273643587 for context.

Implement changes made in https://github.com/twosixlabs/armory/pull/1904 for CARLAMOTAdversarialPatchPyTorch

[Feature]: Config output_dir does not behave as expected

2

1. When supplying a value such as `INTL` as `output_dir`, armory outputs the results to `~/.armory/outputs/INTL_/…` as opposed to what would be expected `~/.armory/outputs/INTL//…` 2. When supplying an absolute path...

Automatic GPU selection

1

Currently when gpu is not specified, armory defaults to running the attack on GPU0 and loading some other data onto all other GPUs. It would be useful to be able...

see #1890

Add support for ablation studies.

1

I.e. take in a collection of config files (or optimally a set of learning-rates or base-images or just a generic list of kwargs to modify) and output a single result...

See https://github.com/twosixlabs/armory/pull/1907#discussion_r1152295158 Relied on by config loading when setting `--check` flag and something else

See https://github.com/twosixlabs/armory/pull/1920#discussion_r1185951735

Right now it is very difficult to understand what the shapes of each dataloader/scenario/attack/exporter expect. This PR hopes to increase the verbosity of our documentation here - particularly to clarify...