- Adopt new layers of EfficientNet-B1 to extract feature maps. The accuracy of small face detection increses obviously.
- Add report and slides
A PyTorch Implementation of Dual Shot Face Detector
We use basenet VGG16, EfficientNet-B0 and EfficientNet-B1 to train DSFD, the model can be downloaded in DSFD with VGG.
The AP using VGG16 in AFW,PASCAL,FDDB as following:
AFW | PASCAL | FDDB |
---|---|---|
99.89 | 99.11 | 98.3 |
The performance on small face detection using EfficientNet-B0 and EfficientNet-B1 is not good enough, we are still working on it.
- pytorch 0.3
- opencv
- numpy
- easydict
- download WIDER face dataset
- modify data/config.py
python prepare_wider_data.py
python train.py --batch_size 4
--model vgg\efficient_b0\efficient_b1
--lr 5e-4
According to yourself dataset path,modify data/config.py
- Evaluate on AFW.
python tools/afw_test.py
- Evaluate on FDDB
python tools/fddb_test.py
- Evaluate on PASCAL face
python tools/pascal_test.py
- test on WIDER FACE
python tools/wider_test.py
you can test yourself image
python demo.py
This page (code, report and presentation) is the group C submission for Final project for CS256: Selected Topic in Artificial Intelligence, Section 2. Leb by Instructor: Mashhour Solh, Ph.D. The group members are:. "Jianwei Li, Peilu Liu, Ruchira Gothankar, Akash Budholia, Akshar Panchal".
The code maybe used for educational and commercial use under no warranties. For questions on this project and code please reach out to: "[email protected]"
This project was conducted with free credits provided by AWS educate team.