/forensic_footprint_detection

Primary LanguagePython

data pre-process

Detection procedure

  1. train Feature extractor: Input images(train+reference) -> AE -> output images(input images) | get features

  2. train Classifier: input image -> trained feature extractor -> output features ->train: 1. CNN classifier -> image labels OR 2. kNN classifier

current progress

Extractor (Tong)

  1. input random (need to load large image files)
  2. auto-encoder (baisc structure tested) : load batched data, save model, load model,prediction (tested)
  3. get encoder output(feature extractor tested): extract encoder part of AE, encoder prediction(feature extraction)(tested)

Classifier: Classifier (haven't started)

current problems

  1. Load large number of images: could solve by: shuffle img paths and load imgs accordingly, need to let keras keep training (Tong)