Recognition of individuals in video surveillance environments

Configuration and execution

Datasets

First you need to download the Casia-Webface for trainning and LFW-funneled for validation and testing and icbrw to test in surveillance environments.

• Casia-Webface: https://paperswithcode.com/dataset/casia-webface
• LFW-funneled: http://vis-www.cs.umass.edu/lfw/
• ICB-RW: http://icbrw.di.ubi.pt/

Face Alignment

To obtain better results, we need to align the faces. You can align all datasets using align_dataset_mtcnn.py (from https://github.com/davidsandberg/facenet). Unfortunately, the tensorflow version was outdated so the following minor changes had to be made in align_dataset_mtcnn.py:

remove the value, feed_in, dim = (inp, input_shape[-1])

#need to import the following packages
import tensorflow.compat.v1 as tf #ignore the red warning
import imageio # to replace misc.imread() as format: imageio.imread(os.path.join(emb_dir, i))
from PIL import Image # to replace misc.resize as format: scaled = np.array(Image.fromarray(cropped).resize((args.image_size, args.image_size), Image.BILINEAR))
#add this things to the code

In the script detect_face.py change:

Row 85, replace as format: data_dict = np.load(data_path, encoding='latin1', allow_pickle=True).item()

Finally, you need to change the path of the directories in run_align.sh and then run the script.

Split the train dataset in train and test

At this stage, you need to split the training dataset in 80% of the photos to train and 20% for validation. You can use create_train_test_splits.py to do that.

The config file

The config.yml file contains all the configurations necessary to train, test, prepare the data, to create the pairs for the ICB-RW and the paths to use to do evaluation.

Train

To train the model you need to change the following paths according to the location of your files:

• train-path;
• validation-path.

If you want to run the custom callback for lfw in the end of each epoch change the following parameters:

• lfw-callback;
• lfw-callback-pairs.

If you don´t want to do this custom callback you have to change the following lines in train.py.

Comment line 81
Change callbacks at line 88 to: callbacks=[modelCheckpoint_callback]

Finally, it's possible to run the file as python3 train.py

Other notes

• Baseline: Train with no align dataset and test with no align dataset and without L2 normalization;
• Align: Train with align dataset and test with align dataset
• Align-L2: Train with align dataset. Test with align dataset, using L2 normalization for the features and calculate euclidean distance between the features
• Align-L2-CosineSim: Train with align dataset. Test with align dataset, using L2 normalization for the features and calculate cosine similarity.