|
--- |
|
license: other |
|
tags: |
|
- computer_vision |
|
- pose_estimation |
|
--- |
|
|
|
|
|
# Model description |
|
|
|
SuperAnimal-Quadruped is a model based on DLCRNet-ms5 (https://www.nature.com/articles/s41592-022-01443-0). The model has been trained with 40K quadruped images from publicly available quadruped images. |
|
|
|
Based on panoptic pose estimation formulation and rich training datasets, the model is able to predict many more keypoints than common human keypoints. |
|
|
|
Copyright 2021-2023 by Mackenzie Mathis, Alexander Mathis, Shaokai Ye and contributors. All rights reserved. |
|
|
|
- Non-commercial use only is permitted |
|
- please cite Ye et al if you use this model in your work https://arxiv.org/abs/2203.07436v1 |
|
- If this license is not suitable for your business or project |
|
please contact EPFL-TTO (https://tto.epfl.ch/) for a full commercial license. |
|
|
|
This software may not be used to harm any animal deliberately. |
|
|
|
# Limitations and bias |
|
|
|
Because most quadruped in images appear in side-view and front-view, the model might have performance degradation in the back-view due to lack of training data. |
|
|
|
The model can have sub-optimal performance if the image is not resized properly as the model is fit to a certain image size (empirically, smaller than 1500 pixels). |
|
|
|
|
|
|
