This code repository includes the source code for the CS330 Final Project Fall 2022 by
Li-Heng Lin| Tz-Wei Mo| Annie Ho|
This repo includes modification to the Multi-LeNet network to include FiLM layers. Multi-VGG is also added to further investigate the effects of FiLM layers by adding task specific batch normalization. CIFAR-10/SVHN dataset is also combined for a new experiment setup for Multi-VGG.
Code for adding FiLM layers in ResNet is also included.
The code uses the following Python packages and they are required: tensorboardX, pytorch, click, numpy, torchvision, tqdm, scipy, Pillow, imageio
We adapt and use some code snippets from:
- https://github.com/isl-org/MultiObjectiveOptimization [Scalarization Multi-task Training]
- https://github.com/kkweon/mnist-competition/blob/master/vgg5.py?fbclid=IwAR1LeFSiJ7ziHQyzDkaHLKVJmDKJw_Z_G4xLJ6hAsaB3PkjqbH0NIqZ52Ao [VGG]
The code base uses configs.json for the global configurations like dataset directories, etc.. Experiment specific parameters are provided seperately as a json file. See the sample.json for an example.
To train a model, use the command:
python multi_task/train_multi_task_scalarization.py --param_file=./sample.json