Big GANs Are Watching You

This repository is the official implementation of Big GANs Are Watching You: Towards Unsupervised Object Segmentation with Off-the-Shelf Generative Models by Andrey Voynov, Stanislav Morozov, and Artem Babenko.
Schematic representation of our approach.

Core steps:

1. Find the BigBiGAN latent direction responsible for background darkening and foreground lightening;
2. Sample real images embeddings and generate segmentation masks with their shifts;
3. Train U-net with that synthetic data.

example of variation along the latent direction

Requirements

python 3.6 or later

torch>=1.4
torchvision>=0.5
tensorboardX
scikit-image
scipy
h5py

at least 12Gb of GPU-memory for batch size 95 (though smaller batch gives competitive result)

For results visualization:

jupyter

Optional for BigBiGAN tf to torch conversion and BigBiGAN embeddings evaluation:

tensorflow_gpu==1.15.2
tensorflow_hub
parse

if troubles – check the authors packages versions in requirements.txt

Training

First run download_all.sh
this scrips downloads:

• BigBiGAN weights of the original tf-hub model (https://tfhub.dev/deepmind/bigbigan-resnet50/1) ported to pytorch;
• precalculated BigBiGAN-embeddings for some popular datasets;
• background darkening and foreground lightening latent directions coefficients.

note that the ImageNet-BigBiGAN-embeddings takes 551Mb.

To train the U-net segmentation model, run this command:

python train_segmentation.py \
    --out "results" \
    --gan_weights "BigGAN/weights/BigBiGAN_x1.pth" \
    --z "embeddings/BigBiGAN_ImageNet_z.npy"
    --bg_direction "BigGAN/weights/bg_direction.pth" \
    --val_images_dir __path_to_images_dir__ \
    --val_masks_dir __path_to_masks_dir__ \

Custom Data

Once you want to use your own images BigBiGAN-embeddings, please run bigbigan_embeddings.ipynb notebook and provide your own data path.

Evaluation

To check the synthetic data / segmentation samples / model metrics, please run inspection.ipynb notebook.

Pre-trained Models

You can download pretrained models here:

• U-net trained on ImageNet-BigBiGAN embeddings using default parameters.
• U-net trained on CUB-200-2011-BigBiGAN embeddings using default parameters.
• U-net trained on Flowers-BigBiGAN embeddings using default parameters.

Results

Our model achieves state-of-the-art results among fully unsupervised methods on many datasets. Check Tables 1, 2, 3 of the original paper for more details. In the tables below we refer our approach as BigBiGAN and E-BigBiGAN.

Object Detection

Saliency Detection

Here are some samples of the model performing on DUTS dataset:

BigBiGAN weights conversion

Once the original BigBiGAN weigths are distributed within tfhub, we also provide the conversion code. See bigbigan2biggan.ipynb. This code is based on this script: https://github.com/ajbrock/BigGAN-PyTorch/blob/master/TFHub/converter.py . Note that BigBiGAN and BigGAN almost share the architecture and we use the author's officially unofficial PyTorch BigGAN implementation.

Licenses

• Unsupervised Discovery of Latent Directions: https://github.com/anvoynov/GANLatentDiscovery
• BigBiGAN weights: https://tfhub.dev/deepmind/bigbigan-resnet50/1
• BigGAN pytorch: https://github.com/ajbrock/BigGAN-PyTorch
• U-Net model code is based on: https://github.com/milesial/Pytorch-UNet
• prefetch_generator: https://github.com/justheuristic/prefetch_generator