Cats Feature Points Estimation

Predicting cats feature points using a feature pyramid network

Kaggle Dataset

Understanding Data

Each record contains an image and a file with 19 values.

The first value represents the number of points (which is 9)

Each two next values represent a coordinate in the picture which shows a feature point:

• Left Eye

• Right Eye

• Mouth

• Left Ear 1

• Left Ear 2

• Left Ear 3

• Right Ear 1

• Right Ear 2

• Right Ear 3

• 

Each image has been resized to 224 * 224

Modeling

• MobileNetV2 as feature extractor
• Feature pyramid with 4 levels
• Detection head with 18 outputs (each corresponding to a point)

Evaluation

•