Predicting cats feature points using a feature pyramid network
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
- MobileNetV2 as feature extractor
- Feature pyramid with 4 levels
- Detection head with 18 outputs (each corresponding to a point)
