Diverse Semantic Image Synthesis via Probability Distribution Modeling (CVPR 2021)

Paper

Zhentao Tan, Menglei Chai, Dongdong Chen, Jing Liao, Qi Chu, [Bin Liu], Gang Hua, Nenghai Yu

Abstract

Semantic image synthesis, translating semantic layouts to photo-realistic images, is a one-to-many mapping problem. Though impressive progress has been recently made, diverse semantic synthesis that can efficiently produce semantic-level multimodal results, still remains a challenge. In this paper, we propose a novel diverse semantic image synthesis framework from the perspective of semantic class distributions, which naturally supports diverse generation at semantic or even instance level. We achieve this by modeling class-level conditional modulation parameters as continuous probability distributions instead of discrete values, and sampling per-instance modulation parameters through instance-adaptive stochastic sampling that is consistent across the network. Moreover, we propose prior noise remapping, through linear perturbation parameters encoded from paired references, to facilitate supervised training and exemplar-based instance style control at test time. Extensive experiments on multiple datasets show that our method can achieve superior diversity and comparable quality compared to state-of-the-art methods.

Installation

Clone this repo.

git clone https://github.com/tzt101/INADE.git
cd INADE/

This code requires PyTorch 1.6 and python 3+. Please install dependencies by

pip install -r requirements.txt

Dataset Preparation

The Cityscapes and ADE20K dataset can be downloaded and prepared following SPADE. The CelebAMask-HQ can be downloaded from CelebAMask-HQ, you need to to integrate the separated annotations into an image file (the format like other datasets, e.g. Cityscapes and ADE20K). The DeepFashion can be downloaded from SMIS, and the version with two persons can be downloaded from OneDrive.

To make or reid the instance map, you can use the following commands:

python make_instances.py --path [Path_to_dataset] --dataset [ade20k | cityscapes | celeba | deepfashion]

Generating Images Using Pretrained Model

Once the dataset is ready, the result images can be generated using pretrained models.

1. Download the pretrained models from the OneDrive, save it in checkpoints/. The structure is as follows:

./checkpoints/
    inade_ade20k/
        best_net_G.pth
        best_net_IE.pth
    inade_celeba/
        best_net_G.pth
        best_net_IE.pth
    inade_cityscapes/
        best_net_G.pth
        best_net_IE.pth
    inade_deepfashion/
        best_net_G.pth
        best_net_IE.pth

The noise_nc is 64 for all pretrained models except which on deepfashion (set to 8). Because we find that it's enough for quality and diversity.

2. Generate the images on the test dataset.

python test.py --name [model_name] --norm_mode inade --batchSize 1 --gpu_ids 0 --which_epoch best --dataset_mode [dataset] --dataroot [Path_to_dataset]

[model_name] is the directory name of the checkpoint file downloaded in Step 1, such as inade_ade20k and inade_cityscapes. [dataset] can be on of ade20k, celeba, cityscapes and deepfashion. [Path_to_dataset] is the path to the dataset. If you want to use encoder, you can add the another option --use_vae.

Training New Models

You can train your own model with the following command:

# To train CLADE and CLADE-ICPE.
python train.py --name [experiment_name] --dataset_mode [dataset] --norm_mode inade --use_vae --dataroot [Path_to_dataset]

If you want to test the model during the training step, please set --train_eval. By default, the model every 10 epoch will be test in terms of FID. Finally, the model with best FID score will be saved as best_net_G.pth.

Calculate FID

We provide the code to calculate the FID which is based on rpo. We have pre-calculated the distribution of real images (all images are resized to 256×256 except cityscapes is 512×256) in training set of each dataset and saved them in ./datasets/train_mu_si/. You can run the following command:

python fid_score.py [Path_to_real_image] [Path_to_fake_image] --batch-size 1 --gpu 0 --load_np_name [dataset] --resize_size [Size]

The provided [dataset] are: ade20k, celeba, cityscapes, coco and deepfashion. You can save the new dataset by replacing --load_np_name [dataset] with --save_np_name [dataset].

New Useful Options

The new options are as follows:

• --use_amp: if specified, use AMP training mode.
• --train_eval: if sepcified, evaluate the model during training.
• --eval_dims: the default setting is 2048, Dimensionality of Inception features to use.
• --eval_epoch_freq: the default setting is 10, frequency of calculate fid score at the end of epochs.

Code Structure

• train.py, test.py: the entry point for training and testing.
• trainers/pix2pix_trainer.py: harnesses and reports the progress of training.
• models/pix2pix_model.py: creates the networks, and compute the losses
• models/networks/: defines the architecture of all models
• 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.

Citation

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

@inproceedings{tan2021diverse,
  title={Diverse Semantic Image Synthesis via Probability Distribution Modeling},
  author={Tan, Zhentao and Chai, Menglei and Chen, Dongdong and Liao, Jing and Chu, Qi and Liu, Bin and Hua, Gang and Yu, Nenghai},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={7962--7971},
  year={2021}
}

Acknowledgments

This code borrows heavily from SPADE.