This model is a version of Yolo v8 nano fine-tuned on the freeclimbs v2 dataset to detect climbing holds, particularly holds on home climbing and "spray" walls. (The dataset is not currently available but I plan to release it in the future.)
It expects a 2560x2560 image (if using the ultralytics library as shown below, it will handle this) and detects a single class - climbing holds.
Usage
from ultralytics import YOLO
model = YOLO("yolov8n-freeclimbs-detect-2.pt")
results = model(
["climbing-wall.jpg"],
imgsz=2560,
max_det=2000)
Performance
| Precision | 0.961 |
| Recall | 0.942 |
| mAP50 | 0.988 |
| mAP50-95 | 0.889 |
(on freeclimbs v2 test set)
License
Copyright (c) 2024 John LaRocque
See LICENSE for license (AGPL 3). Note that an earlier version of this repository erroneously included an MIT license - since this model was fine-tuned from a model licensed under the AGPL 3, which is incompatible with other licenses, I am not actually able to offer that license.
Inference Providers
NEW
This model is not currently available via any of the supported third-party Inference Providers, and
HF Inference API was unable to determine this model's library.