ATTENTION: IF YOU HAVE DOWNLOADED MODELS BEFORE ALREADY - RE-DOWNLOAD THEM PLEASE. I SHOULD HAVE UPDATED THEM WITH NO DILL VERSION(EXCEPT MODEL THAT WAS CHOSEN AS NEW VERSION UPDATE, SO YOU CAN RECEIVE NOTIFICATIONS.<h1 id="if-you-are-having-issues:-cc2d7ef23">If you are having issues:</h1><ol><li>Activate your venv</li><li><code>pip install dill</code></li><li><code>pip install --upgrade ultralytics</code></li></ol>This should fix generic problems.<hr />Small segmentation model aiming to help with head inpainting/enhance. Note: It is not very accurate, and will struggle with long hair.This version is smallest available, compromise in accuracy, but it's just 6mb. Medium(50mb) brings in up a notch, <s>but don't think i will be releasing that one in particular.</s><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/183b086b-17b3-419c-9f33-fb6f243fc981/width=525/183b086b-17b3-419c-9f33-fb6f243fc981.jpeg" />Oke, medium is up.Here are validation results on ~470 images:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/10867a57-5715-46d2-842a-577a258da5b3/width=525/10867a57-5715-46d2-842a-577a258da5b3.jpeg" />Can also be downloaded at HF(whenever i can actually log in): <a target="_blank" rel="ugc" href="https://huggingface.co/Anzhc/Anzhcs_YOLOs">https://huggingface.co/Anzhc/Anzhcs_YOLOs</a>Script im using: <a target="_blank" rel="ugc" href="https://github.com/Anzhc/Training-script-for-Ultralytics-YOLO">https://github.com/Anzhc/Training-script-for-Ultralytics-YOLO</a><hr /><h1 id="usage-xcdda5khq">Usage</h1>Put into `automatic1111/models/adetailer`, or any other place that you know will work with YOLOv8 models.Use.<hr /><h1 id="what-it-does-and-what-is-it-for-trng588nt">What it does and what is it for?</h1>Head segmentation, duh.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/73578f5c-9338-496d-925f-f66848c838e5/width=525/73578f5c-9338-496d-925f-f66848c838e5.jpeg" />It tries to select head, including hair, while avoiding rest. It is hard to avoid rest... Model is very inconsistent with long hair. Overall it should be able to work with wide variety of styles and mediums, as i tries to create a robust enough dataset with quite a bit of abstract, traditional, real and digital content. It is not a big one though, so there is a large room for improvement.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/091ab481-9dca-4dbf-85c1-dded751d9a06/width=525/091ab481-9dca-4dbf-85c1-dded751d9a06.jpeg" />But what you might need it for? To have some fun, while reducing tedium a bit.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/c95f5e18-3242-412b-a425-de12f0dbcfc2/width=525/c95f5e18-3242-412b-a425-de12f0dbcfc2.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3bcfcffc-ddfa-4c69-a6ec-ec3ec41dfd92/width=525/3bcfcffc-ddfa-4c69-a6ec-ec3ec41dfd92.jpeg" />P.S. Those inpaints were done on older version, which had a bit worse masking. All inpaints were done directly with adetailer in single pass with inpaint model.And it took a few tries, hard to get 7 inpaints right in a row xd

Files with SHA256: 3ce0b59880fbdeea494f2842c028e3f961949c97900ec17a606b9b8d1c23dda2