We provide a PyTorch implementation of our contrastive feature loss presented in:

Contrastive Feature Loss for Image Prediction Alex Andonian, Taesung Park, Bryan Russell, Phillip Isola, Jun-Yan Zhu, Richard Zhang

Presented in AIM Workshop at ICCV 2021

• Linux or macOS
• Python 3.6+
• CPU or NVIDIA GPU + CUDA CuDNN

Table of Contents:

1. Setup
2. Dataset Preprocessing
3. Training
4. Evaluating and Visualizing

• Clone this repo:

git clone https://github.com/alexandonian/contrastive-feature-loss.git
cd contrastive-feature-loss

• Create python virtual environment

• Install a recent version of PyTorch and other dependencies specified below.

We highly recommend that you install additional dependencies in an isolated python virtual environment (of your choosing). For Conda+pip users, you can create a new conda environment and then pip install dependencies with the following snippet:

ENV_NAME=contrastive-feature-loss
conda create --name $ENV_NAME python=3.8
conda activate $ENV_NAME
pip install -r requirements.txt

Alternatively, you can create a new Conda environment in one command using conda env create -f environment.yml, followed by conda activate contrastive-feature-loss to activate the environment.

This code also requires the Synchronized-BatchNorm-PyTorch rep.

cd models/networks/
git clone https://github.com/vacancy/Synchronized-BatchNorm-PyTorch
cp -rf Synchronized-BatchNorm-PyTorch/sync_batchnorm .
cd ../../

For COCO-Stuff, Cityscapes or ADE20K, the datasets must be downloaded beforehand. Please download them on the respective webpages. In the case of COCO-stuff, we put a few sample images in this code repo.

Preparing COCO-Stuff Dataset. The dataset can be downloaded here. In particular, you will need to download train2017.zip, val2017.zip, stuffthingmaps_trainval2017.zip, and annotations_trainval2017.zip. The images, labels, and instance maps should be arranged in the same directory structure as in datasets/coco_stuff/. In particular, we used an instance map that combines both the boundaries of "things instance map" and "stuff label map". To do this, we used a simple script datasets/coco_generate_instance_map.py. Please install pycocotools using pip install pycocotools and refer to the script to generate instance maps.

Preparing ADE20K Dataset. The dataset can be downloaded here, which is from MIT Scene Parsing BenchMark. After unzipping the datgaset, put the jpg image files ADEChallengeData2016/images/ and png label files ADEChallengeData2016/annotatoins/ in the same directory.

There are different modes to load images by specifying --preprocess_mode along with --load_size. --crop_size. There are options such as resize_and_crop, which resizes the images into square images of side length load_size and randomly crops to crop_size. scale_shortside_and_crop scales the image to have a short side of length load_size and crops to crop_size x crop_size square. To see all modes, please use python train.py --help and take a look at data/base_dataset.py. By default at the training phase, the images are randomly flipped horizontally. To prevent this use --no_flip.

Models can be trained with the following steps.

1. Ensure you have prepared the dataset of interest using the instruction above. To train on the one of the datasets listed above, you can download the datasets and use --dataset_mode option, which will choose which subclass of BaseDataset is loaded. For custom datasets, the easiest way is to use ./data/custom_dataset.py by specifying the option --dataset_mode custom, along with --label_dir [path_to_labels] --image_dir [path_to_images]. You also need to specify options such as --label_nc for the number of label classes in the dataset, --contain_dontcare_label to specify whether it has an unknown label, or --no_instance to denote the dataset doesn't have instance maps.

2. Run the training script with the following command:

# To train on the COCO dataset, for example.
python train.py --name [experiment_name] --dataset_mode coco --dataroot [path_to_coco_dataset]

# To train on your own custom dataset
python train.py --name [experiment_name] --dataset_mode custom --label_dir [path_to_labels] --image_dir [path_to_images] --label_nc [num_labels]

There are many options you can specify. Please use python train.py --help. The specified options are printed to the console. To specify the number of GPUs to utilize, use --gpu_ids. If you want to use the second and third GPUs for example, use --gpu_ids 1,2.

The training logs are stored in json format in [checkpoints_dir]/[experiment_name]/log.json, with sample generations populated in [checkpoint_dir]/[experiment_name]/web. We provide support for both Tensorboard (with --tf_log) and Weights & Biases (with --use_wandb) experiment tracking.

In order to evaluate and visualize the generations of a trained model, run test.py in a similar manner, specifying the name of the experiment, the dataset and its path:

python test.py --name [name_of_experiment] --dataset_mode [dataset_mode] --dataroot [path_to_dataset]

where [name_of_experiment] is the directory name of the checkpoint created during training. If you are running on CPU mode, append --gpu_ids -1.

Use --results_dir to specify the output directory. --num_test will specify the maximum number of images to generate. By default, it loads the latest checkpoint. It can be changed using --which_epoch.

• train.py, test.py: the entry point for training and testing.
• trainers/contrastive_pix2pix_trainer.py: harnesses and reports the progress of training of contrastive model.
• models/contrastive_pix2pix_model.py: creates the networks, and compute the losses
• models/networks/: defines the architecture of all models
• models/networks/loss: contains proposed PatchNCE loss
• options/: creates option lists using argparse package. More individuals are dynamically added in other files as well. Please see the section below.
• data/: defines the class for loading images and label maps.

This code repo contains many options. Some options belong to only one specific model, and some options have different default values depending on other options. To address this, the BaseOption class dynamically loads and sets options depending on what model, network, and datasets are used. This is done by calling the static method modify_commandline_options of various classes. It takes in theparser of argparse package and modifies the list of options. For example, since COCO-stuff dataset contains a special label "unknown", when COCO-stuff dataset is used, it sets --contain_dontcare_label automatically at data/coco_dataset.py. You can take a look at def gather_options() of options/base_options.py, or models/network/__init__.py to get a sense of how this works.

If you use this code for your research, please cite our paper.

@inproceedings{andonian2021contrastive,
  title={Contrastive Feature Loss for Image Prediction},
  author={Andonian, Alex and Park, Taesung and Russell, Bryan and Isola, Phillip and Zhu, Jun-Yan and Zhang, Richard},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={1934--1943},
  year={2021}
}

This code borrows heavily from pix2pixHD and SPADE and CUT. We thank Jiayuan Mao for his Synchronized Batch Normalization code.