ADetailer YOLOv8

Bingsu/adetailer is a collection of YOLOv8 and YOLOv9 object detection and instance segmentation models specialized for anime and realistic imagery, targeting faces, hands, persons, and clothing. The models are trained on curated datasets including Anime Face CreateML, WIDER Face, AnHDet, COCO2017 (person class), AniSeg, skytnt/anime-segmentation, and DeepFashion2. Key strengths include high mean average precision (mAP) across multiple model sizes and targets. For face detection, the face_yolov9c.pt model achieves a mAP50 of 0.748 and mAP50-95 of 0.433. For hand detection, hand_yolov9c.pt reaches a mAP50 of 0.810 and mAP50-95 of 0.550. Person segmentation models (e.g., person_yolov8m-seg.pt) achieve a mAP50 of 0.849 for bounding boxes and 0.831 for masks. The deepfashion2_yolov8s-seg.pt model for clothing segmentation attains a mAP50 of 0.849 (bbox) and 0.840 (mask), with a mAP50-95 of 0.763 (bbox) and 0.675 (mask). The models are trained on diverse datasets: face models use Anime Face CreateML and WIDER Face; hand models use AnHDet and hand-detection-fuao9; person models use COCO2017, AniSeg, and skytnt/anime-segmentation (a ~18GB dataset of background, foreground, and mask images under CC0-1.0); clothing models use DeepFashion2. The skytnt/anime-segmentation dataset includes 8,057 background images, 11,802 foreground images, and 1,111 real images with masks, sourced from Danbooru and AniSeg, with backgrounds restored via Real-ESRGAN and cleaned using DeepDanbooru plus manual review. The following table summarizes available model variants and their benchmark performance:

Model	Target	mAP50	mAP50-95
face_yolov8n.pt	2D / realistic face	0.660	0.366
face_yolov8n_v2.pt	2D / realistic face	0.669	0.372
face_yolov8s.pt	2D / realistic face	0.713	0.404
face_yolov8m.pt	2D / realistic face	0.737	0.424
face_yolov9c.pt	2D / realistic face	0.748	0.433
hand_yolov8n.pt	2D / realistic hand	0.767	0.505
hand_yolov8s.pt	2D / realistic hand	0.794	0.527
hand_yolov9c.pt	2D / realistic hand	0.810	0.550
person_yolov8n-seg.pt	2D / realistic person	0.782 (bbox) / 0.761 (mask)	0.555 (bbox) / 0.460 (mask)
person_yolov8s-seg.pt	2D / realistic person	0.824 (bbox) / 0.809 (mask)	0.605 (bbox) / 0.508 (mask)
person_yolov8m-seg.pt	2D / realistic person	0.849 (bbox) / 0.831 (mask)	0.636 (bbox) / 0.533 (mask)
deepfashion2_yolov8s-seg.pt	realistic clothes	0.849 (bbox) / 0.840 (mask)	0.763 (bbox) / 0.675 (mask)

The deepfashion2 model supports 13 clothing categories including short_sleeved_shirt, trousers, skirt, and dress variants. All models are created using the official ultralytics library and are hosted as managed APIs on gigarouter.