Data record
Model training data
The v1 artifact is an upstream pretrained YOLO26x-Pose model from Ultralytics. The local repository does not contain a local training or fine-tuning run, and no local training dataset was used to produce the published checkpoint.
- Data collector / dataset owner: Ultralytics
- Dataset: COCO Keypoints
- Dataset version or snapshot: Not recorded in this inventory entry
- Annotation format: COCO Keypoints / human pose annotations
- Curation notes: The packaged model is an upstream pretrained pose-estimation artifact. No local training data has been added to this repository.
- Train/validation/test split records: The upstream training split details are not reproduced in this inventory entry.
- Class distribution: 1 class (
person) - Keypoints: 17 human pose keypoints
- Output features: 51 per detected person (17 keypoints × 3) plus bounding box
(x1, y1, x2, y2)
Trainer benchmark data
The published YOLO26x-Pose benchmark is an upstream Ultralytics reference benchmark. The local repository does not claim to have reproduced the upstream training benchmark.
| Model | Input | Dataset | mAP50-95 | mAP50 |
|---|---|---|---|---|
| YOLO26x-Pose | 960 × 960 | COCO Keypoints | 71.6% | 91.6% |
The reference benchmark represents the published performance of the pretrained YOLO26x-Pose model. The documented values are kept separate from any future independent validation performed by the local validator.
The exact upstream training-data release details, training split counts, benchmark hardware configuration, and training run configuration are not recorded in this inventory entry.
Independent validation data
Vivek must provide a hold-off dataset that is independent of the upstream reference benchmark before recording independent validation results. The validator should record the dataset source, version or capture window, image count, labeled person count, annotation format, evaluation method, and any permissions required to run the evaluation.
The hold-off data must remain outside this repository unless its owner has approved redistribution.
The supplied evaluation notebook expects the following local structure:
evaluation_data/
├── images/
│ ├── image_001.jpg
│ ├── image_002.jpg
│ └── ...
├── labels/
│ ├── image_001.txt
│ ├── image_002.txt
│ └── ...
└── data.yaml