1. Landmark localization

1-1. What is landmark localization

• Landmark localization : predicting the coordinates of keypoints

1-2. Coordinate regression VS heatmap classification

1. coordinate regression : Usually inaccurate and biased

2. Heatmap classification : better performance but high computational cost

1-3. Hourglass network

• Stacked hourglass modules allow for repeated bottom-up and top-down inference that refines the output of the previous hourglass module

1-4. DensePose

• UV map is a flattened representation of 3D geometry
• canonical coordinate : UV map is invariant to motion
• DensePose R-CNN : Faster R-CNN + 3D surface regression branch

1-5. RetinaFace

• RetinaFace : FPN + Multi-task branches