https://github.com/haydenshively/fddb-unet

an attempt at using the UNet architecture to detect faces in FDDB images
https://github.com/haydenshively/fddb-unet

face-detection keras opencv unet

Last synced: 23 days ago
JSON representation

an attempt at using the UNet architecture to detect faces in FDDB images

• Host: GitHub
• URL: https://github.com/haydenshively/fddb-unet
• Owner: haydenshively
• Created: 2017-11-24T15:36:58.000Z (almost 7 years ago)
• Default Branch: master
• Last Pushed: 2019-02-03T00:13:08.000Z (almost 6 years ago)
• Last Synced: 2023-10-20T20:10:41.800Z (about 1 year ago)
• Topics: face-detection, keras, opencv, unet
• Language: Python
• Homepage:
• Size: 10.7 KB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# FDDB UNet

This project represents my first foray into the world of neural networks. Planning to incorporate this code into my Aslan project, I had hoped that the UNet architecture would be able to highlight image regions that contained faces. However, FDDB contained many faces located in the center of the image, and the network learned to highlight just that area. Obviously this wasn't very useful, so I abandoned this project and began work on FDDB-MobileNet. If I ever come back to it, the first thing I'll do is implement dataset augmentation, which may solve the center-highlighting issue.  

  

**FDDBToSamples.py**:  

converts the raw FDDB files into more convenient numpy files  

  

**unet.py**:  

defines a UNet architecture in terms of Keras' "Sequential Model" API  

  

**train.py**:  

trains a model using the numpy files and UNet architecture from the code above  

  

**predict.py**:  

runs the trained model on a portion of the dataset and saves results to a file  

  

**see_results.py**:  

displays the saved results  

  

## Prerequisites

Keras, OpenCV, numpy  

Dataset: http://vis-www.cs.umass.edu/fddb/