cuimengjie / siamesenetwork-pytorch Goto Github PK

• The prototype is still buggy and need to be fixed.
• use command python main.py to run face recognition with siamese network prototype.
• Dataset available from https://github.com/StephenMilborrow/muct.git
• New dataset available from http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html
• Need to create a trainset/testset/validset csv/txt file before training which include names of all images.
• File format as IMAGE_NAME IMAGE_LABEL in each line
• Network contains 11 conv + 4 pooling + 2 maxout + 2 fc layers.

If program crashes, try to reduce BATCH_SIZE, require roughly 11G memory for BATCH_SIZE=130.

• Add validation step
• Improving algorithm efficiency
• Maximize GPU usage
• Add comment
• Add functionality to save and load trained model
• Rewrite part of code for easy customizing
• Rewrite according to given API

• python 3.*
• pytorch 0.4.1
• torchvision
• matplotlib
• skimage
• numpy
• pandas

• The origin of this project is from one of my second year module in University of Nottingham.
• That project focus on build a real time face recognition application for low quality video/CCTV.
• This is face recognition part of that origin project, all writen by author.