pytorch-face_recognition

A pytorch implementation of CenterNet based face recognition

Example

cd ./egs/CelebA/unet/

1. Training

You have to set proper path in train.sh for dataset.

image_root="../../../dataset/CelebA/img_celeba"
train_path="../../../dataset/CelebA/annotations/train.txt"
valid_path="../../../dataset/CelebA/annotations/valid.txt"

Then,

. ./train.sh <OUTPUT_DIRECTORY>

If you want to resume training,

. ./train.sh <OUTPUT_DIRECTORY> <MODEL_PATH>

You can change model configuration by changing patameters in train.sh.

Early stopping

In my implementation, if the model doesn't update the best loss (= smallest) in continuous 3 epochs, the learning rate will be halved. if the model doesn't in continuous 10 epochs, the training will be stopped before the training epoch reaches the number you specified.

2. Evaluation

You have to set proper path in eval.sh for dataset.

image_root="../../../dataset/CelebA/img_celeba"
test_path="../../../dataset/CelebA/annotations/test.txt"

Then,

. ./eval.sh <OUTPUT_DIRECTORY> <MODEL_PATH>

You can change model configuration by changing patameters in eval.sh. The parameters shold match configuration of <MODEL_PATH>.

3. Demo

. ./demo.sh