MCNCC

Shoeprint matching algorithm using multiple feature channels from a pretrained neural network (googlenet) and normalized cross correlation

Steps:

• download the FID-300 dataset and create a new project folder containing the python file and the "datasets" folder Link to the FID-300 dataset: https://fid.dmi.unibas.ch/

• (optionally) If you don't want to run through the whole dataset (depending on your gpu, this can take a long time), you can create two additonal folders with subsets or download the subsets from this repository (tracks_cropped_Subset and Subset in datasets/FID-300) in the datasets/FID-300 folder. For example the first 50 track images and the corresponding reference images.

• create a virtual environment for this project and install all the necessary packages. What you need to install:

  • torch (go to https://pytorch.org/ and follow the guide)
  • tqdm
  • matplotlib
  • pandas

• run the program with the following arguments:

optional arguments:

-h, --help (show this help message and exit)

-t TRACKS, --tracks TRACKS (define track folder)

-rf REFS, --refs REFS (define reference folder)

-str STRIDE, --stride STRIDE (stride for convolutions)

-r, --rot (add rotation)

-ris START, --start START (rotation interval start)

-rie END, --end END (rotation interval end)

-sf SCOREFILE, --scorefile SCOREFILE (scorefilename)

-cmc, --cmc (calculate cmc)

-cmcf CMC_FILE, --cmc_file CMC_FILE (cmc filename)

for example

     python3 mcncc.py -t tracks_cropped -rf references

• after running the program, a .npy file is created storing the correlation matrix (rows: number of tracks in the chosen track folder, columns: number of reference images in the chosen reference image folder)

• use the cmc argument in order to create cmc-plots from your correlation score-files

This function creates for example following graphs:

     python drive/My\ Drive/MCNCC/mcncc.py -cmc 

     python drive/My\ Drive/MCNCC/mcncc.py -str 4 -cmc

     python drive/My\ Drive/MCNCC/mcncc.py -cmc -avgp

• If you don't have access to a GPU you can also use google collaboratory through this link in order to test out the algorithm: https://drive.google.com/drive/folders/13txeoZfnQ6rAHktlV3-q9x69nJ-rg8qt?usp=sharing