Training new neural network models

Note: Kenneth Jung noticed that the model definitions are slightly different than the pre-trained models. For more information, see issues #351 and #349.

---

We have also released our deep neural network (DNN) training infrastructure to promote an open ecosystem and enable quicker bootstrapping for new research and development.

There is a distinction between training the DNN model for feature representation and training a model for classifying people with the DNN model. If you're interested in creating a new classifier, see Demo 3. This page is for advanced users interested in training a new DNN model and should be done with large datasets (>500k images) to improve the feature representation.

Warning: Training is computationally and memory expensive and takes a day on our Tesla K40 GPU.

A rough overview of training is:

1. Create raw image directory.

Create a directory for your raw images so that images from different people are in different subdirectories. The names of the labels or images do not matter, and each person can have a different amount of images. The images should be formatted as jpg or png and have a lowercase extension.

$ tree data/mydataset/raw
person-1
├── image-1.jpg
├── image-2.png
...
└── image-p.png

...

person-m
├── image-1.png
├── image-2.jpg
...
└── image-q.png

2. Preprocess the raw images

If you plan to compute LFW accuracies, remove all LFW identities for your dataset. We provide an example script doing this with string matching in remove-lfw-names.py.

Change 8 to however many separate processes you want to run: for N in {1..8}; do ./util/align-dlib.py <path-to-raw-data> align outerEyesAndNose <path-to-aligned-data> --size 96 & done.

Prune out directories with less than 3 images per class with ./util/prune-dataset.py <path-to-aligned-data> --numImagesThreshold 3.

3. Train the model

Run training/main.lua to start training the model. Edit the dataset options in training/opts.lua or pass them as command-line parameters. This will output the loss and in-progress models to training/work. The GPU memory usage is determined by the -peoplePerBatch and -imagesPerPerson parameters, which default to 15 and 20 respectively and consume about 12GB of memory. These determine an upper-bound on the mini-batch size and should be reduced for less GPU memory consumption.

Warning: Metadata about the on-disk data is cached in training/work/trainCache.t7 and assumes the data directory does not change. If your data directory changes, delete these files so they will be regenerated.

Stopping and starting training

Models are saved in the work directory after every epoch. If the training process is killed, it can be resumed from the last saved model with the -retrain option. Also pass a different -manualSeed so a different image sequence is sampled and correctly set -epochNumber.

4. Analyze training

Visualize the loss with training/plot-loss.py. Install the Python dependencies from training/requirements.txt with pip2 install -r requirements.txt.