AnimalPose : Pose Esimation for Animals
This is an introduction to「AnimalPose」, a machine learning model that can be used with ailia SDK. You can easily use this model to create AI applications using ailia SDK as well as many other ready-to-use ailia MODELS.
Overview
AnimalPose takes an image of an animal as input and computes a skeleton made of 20 keypoints. Since the model works on cows, it could for example be used in the field of agriculture.
Architecture
AnimalPose is published as part of mmpose, a general-purpose pose estimation framework. Two pre-trained models of AnimalPose are provided, one using hrnet and the other using pose_resnet.
Both models are based on a Top-down approach, which detects the keypoints of a single animal at a time. Animals are first detected using object detection models such as YOLO, and then the keypoints of each animal are computed by AnimalPose. More details on the difference between Top-down and Bottom-up approaches here.
Dataset
AnimalPose was trained using Animal-Pose Dataset, which has been partially created based on PASCAL2011 dataset annotations and images.
The dataset contains more than 3000 images annotated in five categories, with a total of 5517 instances. Each instance has 20 key points defined: 4 paws, 2 eyes, 2 ears, 4 elbows, nose, throat, withers, tail base, and 4 knees points.
Usage
Use the following command to run AnimalPose on a video file.
$ python3 animalpose.py -v input.mp4
Here is a result example.
Related topics
ax Inc. has developed ailia SDK, which enables cross-platform, GPU-based rapid inference.
ax Inc. provides a wide range of services from consulting and model creation, to the development of AI-based applications and SDKs. Feel free to contact us for any inquiry.