https://github.com/halcy/deepdanbooruactivationmaps

Scripts to calculate interest regions for tags for the DeepDanbooru tagger model
https://github.com/halcy/deepdanbooruactivationmaps

Last synced: 11 months ago
JSON representation

Scripts to calculate interest regions for tags for the DeepDanbooru tagger model

• Host: GitHub
• URL: https://github.com/halcy/deepdanbooruactivationmaps
• Owner: halcy
• Created: 2019-08-15T21:34:13.000Z (almost 7 years ago)
• Default Branch: master
• Last Pushed: 2020-08-20T14:04:23.000Z (almost 6 years ago)
• Last Synced: 2025-03-30T18:43:56.639Z (about 1 year ago)
• Language: Jupyter Notebook
• Size: 10.4 MB
• Stars: 34
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# This repository has been superseded by https://github.com/halcy/AnimeFaceNotebooks - probably have a look at that instead.

# DeepDanbooru Activation Maps

This repository contains scripts that can be used to calculate regions of interest for recognized tags on a given image, using the DeepDanbooru tag estimation model (keras conversion, [weights can be downloaded here](https://t.co/OMRYdpUrw0?amp=1), distributed with permission of the models creator).

Some example outputs:

![example output: open mouth](https://raw.githubusercontent.com/halcy/DeepDanbooruActivationMaps/master/example_open_mouth.png)

![example output: cleavage](https://raw.githubusercontent.com/halcy/DeepDanbooruActivationMaps/master/example_cleavage.png)

![example output: barefoot](https://raw.githubusercontent.com/halcy/DeepDanbooruActivationMaps/master/example_barefoot.png)

One somewhat accidental use of this is automatic censoring of images by blurring out regions that correspond to explicit / questionable tags. Code that does this can be found [in this notebook](https://github.com/halcy/DeepDanbooruActivationMaps/blob/master/DeepDanbooru-ActivationMaps-Censorship.ipynb) (warning: includes output, which is to say, a censored explicit image).

The codes speed has been improved somewhat, but suggestions for how to speed it up more very welcome.