Closed rafaelorozco closed 1 month ago
Base: 88.06% // Head: 35.69% // Decreases project coverage by -52.37%
:warning:
Coverage data is based on head (
6f30f61
) compared to base (6d2fba0
). Patch coverage: 6.15% of modified lines in pull request are covered.:exclamation: Current head 6f30f61 differs from pull request most recent head 36a3688. Consider uploading reports for the commit 36a3688 to get more accurate results
:umbrella: View full report at Codecov.
:loudspeaker: Do you have feedback about the report comment? Let us know in this issue.
This is meant to be a bare-minimum-but-reasonable-results conditional sampling example on images. Runs on CPU in 3 minutes and produces OK results. Users can run for longer epochs and with more training samples to get better results.
Added an even more minimal example to the README so that users can see the basic workflow of training conditional flow and sampling it.
To get things clean, I made some changes to conditional glow. Basically some complexity regarding the reshaping of the conditioning variable is hidden from the user so that sampling looks exactly like the math of C-INN: