After Detailer: Difference between revisions
StableTiger3 (talk | contribs) mNo edit summary |
StableTiger3 (talk | contribs) |
||
| Line 8: | Line 8: | ||
Models: | == Grasping the Nuances of Detection and Mask Parameters == | ||
Don't sweat it if you're new to aDetailer! Though the defaults are pretty good, understanding what each setting does can be a game-changer: | |||
=== Detection Model Confidence Threshold === | |||
This is your go-to setting for deciding the minimum confidence level the model needs to flag something. Looking to capture more faces? Go ahead and lower this threshold (say, to around 0.3). Tweak this to get more or fewer detections as you see fit. | |||
=== Mask Min/Max Area Ratio === | |||
Ever annoyed by tiny, irrelevant objects getting picked up? Adjusting the minimum area ratio can help you weed those out. This setting basically tells the model what size range is cool for masks. | |||
== Diving into Inpainting Settings == | |||
When it comes to inpainting, "Inpaint denoising strength" is your MVP. It controls how much denoising happens during the inpainting process. Tweak it until you like what you see. | |||
In most scenarios, you'll probably want to stick with "Inpaint only masked" if you're inpainting faces. It's generally the way to go. | |||
== Models: | |||
<code>face_yolov8s.pt</code> and <code>face_yolov8n.pt</code> are both models from the <code>adetailer</code> repository on Hugging Face1. Both models are used for 2D realistic face detection. The main difference between the two is that <code>face_yolov8s.pt</code> is more accurate in detecting faces than <code>face_yolov8n.pt</code>. Regarding their performance,<code>face_yolov8s.pt</code> has a higher mean average precision (mAP) of 0.713 at an intersection over union (IoU) threshold of 0.50 and 0.404 at an IoU threshold of 0.50-0.95, while <code>face_yolov8n.pt</code> has a mAP of 0.660 at an IoU threshold of 0.50 and 0.366 at an IoU threshold of 0.50-0.951. | <code>face_yolov8s.pt</code> and <code>face_yolov8n.pt</code> are both models from the <code>adetailer</code> repository on Hugging Face1. Both models are used for 2D realistic face detection. The main difference between the two is that <code>face_yolov8s.pt</code> is more accurate in detecting faces than <code>face_yolov8n.pt</code>. Regarding their performance,<code>face_yolov8s.pt</code> has a higher mean average precision (mAP) of 0.713 at an intersection over union (IoU) threshold of 0.50 and 0.404 at an IoU threshold of 0.50-0.95, while <code>face_yolov8n.pt</code> has a mAP of 0.660 at an IoU threshold of 0.50 and 0.366 at an IoU threshold of 0.50-0.951. | ||
Revision as of 00:27, 17 September 2023
What is adetailer?
After Detailer (also known as adetailer) is an extension for enhancing image details. Most noteably, particular parts of the body. Since we spend a lot of time in our lives looking at the face, it is an area that needs particular attention when generating an image because the viewer is able to pick up on the smallest flaws. No additional downloads are required post-initial installation. The extension features specialized models in three key categories: Face, Hand, and Person.
The extension will automatically detect the body part based on the model that is selected, and then improve that area.
How to get started
If you'd like to get started with adetailer, you'll have to install the extension first. If you'd like more information on how to in
Grasping the Nuances of Detection and Mask Parameters
Don't sweat it if you're new to aDetailer! Though the defaults are pretty good, understanding what each setting does can be a game-changer:
Detection Model Confidence Threshold
This is your go-to setting for deciding the minimum confidence level the model needs to flag something. Looking to capture more faces? Go ahead and lower this threshold (say, to around 0.3). Tweak this to get more or fewer detections as you see fit.
Mask Min/Max Area Ratio
Ever annoyed by tiny, irrelevant objects getting picked up? Adjusting the minimum area ratio can help you weed those out. This setting basically tells the model what size range is cool for masks.
Diving into Inpainting Settings
When it comes to inpainting, "Inpaint denoising strength" is your MVP. It controls how much denoising happens during the inpainting process. Tweak it until you like what you see.
In most scenarios, you'll probably want to stick with "Inpaint only masked" if you're inpainting faces. It's generally the way to go.
== Models:
face_yolov8s.pt and face_yolov8n.pt are both models from the adetailer repository on Hugging Face1. Both models are used for 2D realistic face detection. The main difference between the two is that face_yolov8s.pt is more accurate in detecting faces than face_yolov8n.pt. Regarding their performance,face_yolov8s.pt has a higher mean average precision (mAP) of 0.713 at an intersection over union (IoU) threshold of 0.50 and 0.404 at an IoU threshold of 0.50-0.95, while face_yolov8n.pt has a mAP of 0.660 at an IoU threshold of 0.50 and 0.366 at an IoU threshold of 0.50-0.951.