Deep Regression Forests for Age Estimation

A simple pytorch implementation of DRFs.

Experiments

Best Hyper-parameters:

batch size: 32

base learning rate:0.2

VGG pretrained: VGG-Face

or

batch size: 32

base learning rate:0.05

VGG pretrained: VGG-Face

different learning rate in layers

MAE: 2.4~2.5 CS: 88%

Grad-CAM

tree_id: choose a tree

node_id: choose a node

python grad_cam.py --tree_id=0 --node_id=16

example