Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -5,6 +5,7 @@ pipeline_tag: object-detection
|
|
| 5 |
base_model: Ultralytics/YOLO11
|
| 6 |
datasets:
|
| 7 |
- deepghs/danbooru2024
|
|
|
|
| 8 |
metrics:
|
| 9 |
- mAP
|
| 10 |
tags:
|
|
@@ -211,7 +212,8 @@ Anime images from `deepghs/danbooru2024`, automatically labeled with DWPose whol
|
|
| 211 |
keypoints become padded boxes). v3 re-labels that corpus with v2 and keeps a box only when v2 is confident
|
| 212 |
or it lines up with the DWPose box. Training also mixes in ANFDet (a public-domain anime-foot dataset), a
|
| 213 |
few hundred hand-labeled images, and feet-free images as hard negatives to cut false positives. v3 was
|
| 214 |
-
trained on roughly 286k images.
|
|
|
|
| 215 |
|
| 216 |
## Intended use and limitations
|
| 217 |
|
|
|
|
| 5 |
base_model: Ultralytics/YOLO11
|
| 6 |
datasets:
|
| 7 |
- deepghs/danbooru2024
|
| 8 |
+
- deepghs/danbooru2024-sfw
|
| 9 |
metrics:
|
| 10 |
- mAP
|
| 11 |
tags:
|
|
|
|
| 212 |
keypoints become padded boxes). v3 re-labels that corpus with v2 and keeps a box only when v2 is confident
|
| 213 |
or it lines up with the DWPose box. Training also mixes in ANFDet (a public-domain anime-foot dataset), a
|
| 214 |
few hundred hand-labeled images, and feet-free images as hard negatives to cut false positives. v3 was
|
| 215 |
+
trained on roughly 286k images. The earlier v1 and v2 used the smaller SFW subset
|
| 216 |
+
(`deepghs/danbooru2024-sfw`); v3 moved to the full `deepghs/danbooru2024` corpus.
|
| 217 |
|
| 218 |
## Intended use and limitations
|
| 219 |
|