Next step for series of rips https://nyaa.iss.one/view/1202653 for period 08.2019-01.2020
**This imageset not intended to be "original and complete" but rather "the best of" and "representative"
to help users not to loose interesting fandom, artist or even single prominent picture**
Used sources (relative priorities high to low):
* safebooru.org (ID 29xxxxx)
* yande.re (with some questionable images in Q folders)
* gelbooru.com (a little bit NSFW in Q)
* konachan.com
* anime-pictures.net
* e-shuushuu.net
* chan.sankakucomplex.com (with Q, only from daily top 200)
* danbooru.donmai.us
File names structure : **%website% - %id% - %copyright% ~ %character% (%artist%).%ext%**
Images zipped according source, ID range (not to exceed 2 GB) and aspect ratio (dimensions 2 folders):
- “squares” 1x1 (+/- 20%)
- “pages” 3x4 (+/- 8%)
- "highpages" 2x3 (+/- 40%)
- “screens” 3x2 (+/- 40%)
Transformations and filters:
- Mpixels >= 1.2, width >= 900, height >= 900
- PNG converted to JPG
- downsize to 64MPix and/or maxsize 9000 px
- most explicit (1/3 to 2/3 depending on source) images excluded from questionable
- comic, 4koma, overtexted, primitives cleared
- crops when large simple or dirty background
- sometimes gamma correction and other nontrivial improvements
- carefully deduplicatied (with AntiDupl NET)
After some processing and bundling this release will be a part of [BOORU-CHARS OPEN DATASET](https://nyaa.iss.one/view/1206322) "volume" **V2020A**
with thumbnail (512x512px images) and supporting structured metadata - tags, image statistical properties and results of some processing
Comments - 1
mioscape