metadata
library_name: lightning-pose
tags:
- lightning-pose
- pose-estimation
trained_by: Anna Lakunina
published_by: Hilda Azimi
data_name: DATA_NAME
data_address: he S3 path or data asset ID
backbone: resnet50_animal_ap10k
losses_used:
- pca_singleview
num_keypoints: 18
keypoint_names:
- eye_top
- eye_right
- eye_bottom
- eye_left
- pupil_top
- pupil_right
- pupil_bottom
- pupil_left
- pupil_top_right
- pupil_bottom_right
- pupil_bottom_left
- pupil_top_left
- reflection
- nostril
- nose_tip
- whisker_1
- whisker_2
- whisker_3
num_views: 1
views:
- single
epochs: 300-300
train_prob: 0.95
val_prob: 0.05
latest_version: v1.0
perf_metric: supervised_rmse
perf_value: 0.64
perf_unit: resized-px
perf_set: validation
Lightning Pose Model
Published by Hilda Azimi · Trained by Anna Lakunina on DATA_NAME (he S3 path or data asset ID).
- Backbone:
resnet50_animal_ap10k - Losses used:
['pca_singleview'] - Keypoints (18):
eye_top, eye_right, eye_bottom, eye_left, pupil_top, pupil_right, pupil_bottom, pupil_left, pupil_top_right, pupil_bottom_right, pupil_bottom_left, pupil_top_left, reflection, nostril, nose_tip, whisker_1, whisker_2, whisker_3 - Epochs: 300-300, train/val split: 0.95/0.05
- Supervised RMSE (validation): 0.64 resized-px
Performance
Supervised RMSE (from TensorBoard logs, unit: resized-px - measured on the resized training image, not original-frame pixels). Lower is better.
No per-keypoint pixel-error file was found; the numbers below come from the training logs instead.
Source: events.out.tfevents.1736539402.f2768a15001d.48.0
| Set | Final RMSE | Best RMSE |
|---|---|---|
| train | 0.34 | 0.29 |
| validation | 0.77 | 0.64 |
Changelog
v1.0
what this model is / what changed