https://github.com/vibhujawa/dl_face_detection_siamese_network

Face Detection using Siamese Network
https://github.com/vibhujawa/dl_face_detection_siamese_network

pytorch siamese-network

Last synced: about 2 months ago
JSON representation

Face Detection using Siamese Network

• Host: GitHub
• URL: https://github.com/vibhujawa/dl_face_detection_siamese_network
• Owner: VibhuJawa
• Created: 2017-11-10T10:32:30.000Z (almost 8 years ago)
• Default Branch: master
• Last Pushed: 2018-09-16T01:39:11.000Z (about 7 years ago)
• Last Synced: 2025-03-26T05:11:15.524Z (6 months ago)
• Topics: pytorch, siamese-network
• Language: Jupyter Notebook
• Homepage:
• Size: 1.03 MB
• Stars: 0
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Face Detection using Siamese Network

This repo explores face detection using Siamese Network (One Shot Learning) using contrastive loss. 

## Without Data Augmentation

###	Commands to run: 

    python p1a.py --save  "model_p1a_non_augment" --augment "N"

    

##	Hyper Parameters:

    Epoch=4

    Batch Size=20

    

## Reasoning:

Without augmentation, we suffer from severe overfitting, the training accuracy goes to 99 percent if we keep on increasing epochs while testing accuracy decreases back to 50 percent which is the same as random classification. 

So, I chose less epochs and I got an accuracy of:

  *	53 % on testing images

  *	71 % on training images

##	Hyper Parameters with Data Augmentationg:	

    Epoch=10

    Batch Size=20