FAME-tests

Getting started

Install

Clone this repository:

git clone https://github.com/nvanderperren/FAME-tests.git && cd FAME-tests

Install all neceassary python packages. !IMPORTANT! use Python 3.8, some packages don't support Python 3.9.

pip3 install -r requirements.txt

Prepare dataset (Kunstenpunt only)

The workflow needs a CSV with colums image path (= absolute path of image) and name (name of person on picture or unknown). The script preparations_KP.py creates this CSV for the Kunstenpunt images.

1. mount the hard disk
2. add portret and production folders in the production_dirs and portret_dirs variables in preparations_KP.py
3. start the script: python3 preparations_KP.py

Start the workflow

The workflow needs a CSV with columns image path (= absolute path of image) and name (name of person on picture or unknown).

You can adjust some parameters in workflow.py:

1. treshold: is now 0.7
2. csv_file: is now data/filenames.csv, which is created in the first step

Then, start the script:

python3 workflow.py

(Optional) Prepare data for labeling tool

(dit stuk moet nog beter uitgewerkt worden)

EURECA project is used to validate the results. To set up the labeling tool, some files need to be created. The scripts/prepare_labeling.py script creates these files.

Start the script with python3 scripts/prepare_labeling.py.

You'll see that a data/labeling folder is created in which you can find two CSV files.

Results

You will find:

1. a csv with predictions (predictions.csv) in the data/ folder
2. cropped faces in the data/faces/ folder
3. a visualisation of the clusters in the data/clusters/ folder
4. a UMAP visualisation (UMAP_clusters.html) in the data/ folder
5. files needed for the labeling tool in the data/labeling/ folder:
   1. images.csv: a list with face ID, path of cropped image, predictions and alike faces of each found face
   2. metadata.csv: additional metadata per face (cropped image)