Train Your Own Model From Scratch

Download the training data

For classification and bounding box regression tasks, download WiderFace

For facial landmark regression task, download CelebA

After download these two dataset, you will get file structure like this

WIDER_FACE

├── WIDER_FACE
│   ├── wider_face_split
│   ├── WIDER_test
│   ├── WIDER_train
│   └── WIDER_val

CelebA

├── CelebA
│   ├── Anno
│   ├── Eval
│   ├── img_align_celeba
│   ├── img_celeba
│   └── README.txt

Then, link these folders to "mtcnn/datasets".

ln -s /path/to/WIDER_FACE/* mtcnn/datasets/WIDER_FACE/
ln -s /path/to/CelebA/* mtcnn/datasets/CelebA/

First, we generate training data for pnet.

python scripts/gen_pnet_train.py

Train pnet with epoch 10, batchsize 256 and gpu:0.

python scripts/train_pnet.py -e 10 -b 256 -o output/pnet.torchm -dv cuda:0 -r True

Generate training data for rnet

python scripts/gen_pnet_train.py -m output/pnet.torchm

Train rnet

python scripts/train_rnet.py -e 10 -b 256 -o output/rnet.torchm -dv cuda:0

Generate training data for onet

python scripts/gen_onet_train.py -pm output/pnet.torchm -rm output/rnet.torchm

Train onet

python scripts/train_onet.py -e 9 -b 256 -o output/onet.torchm -dv cuda:1 -r True