Skip to content

Visualize feature maps - #53

Merged
mattersoflight merged 2 commits into
mainfrom
visualize-features
Oct 31, 2023
Merged

mattersoflight merged 2 commits into
mainfrom
visualize-features

Conversation

@ziw-liu

@ziw-liu ziw-liu commented Sep 13, 2023

Copy link
Copy Markdown
Contributor

Add script to visualize feature maps produced by deep layers with PCA and t-SNE.

@ziw-liu
ziw-liu marked this pull request as ready for review September 13, 2023 19:11
@ziw-liu ziw-liu added the enhancement New feature or request label Sep 13, 2023
@mattersoflight
mattersoflight merged commit 794eebc into main Oct 31, 2023
@ziw-liu
ziw-liu deleted the visualize-features branch November 1, 2023 04:46
alxndrkalinin added a commit that referenced this pull request Jul 30, 2026
v0.9.0a1 is cubic's "repository-wide bug sweep" release (PR #53, 84 files,
+9669/-2980). Two fixes drive this bump:

- `ssim` now honours `normalize` instead of letting it vanish into
  `**kwargs` (skimage consumes only K1/K2/sigma/use_sample_covariance and
  drops the rest silently), and rejects keywords skimage would drop, so a
  typo raises instead of disappearing.
- `scale_invariant=True` combined with `normalize`/`normalization` now
  raises rather than returning a quietly wrong number. That combination was
  silently wrong for psnr and nrmse too (psnr +45.027 -> +59.593, nrmse
  0.005606 -> 0.001048).

The sweep also rewrote the spectral/FSC, MicroSSIM, MS-SSIM, feature
(KID/FID) and average_precision modules, so essentially every pixel and
feature metric in the benchmark can move across this pin. It is therefore
paired with a full re-eval rather than bumped on its own.

Pinned to the tag; v0.9.0a1 is not on PyPI yet and this dependency has
always resolved from git.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

enhancement New feature or request

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants