metadata

tags:
  - huggan
  - gan
datasets:
  - arakesh/uavid-15-hq-mixedres
license: mit

MyModelName

Model description

Pix2pix Model is a conditional adversarial networks, a general-purpose solution to image-to-image translation problems. These networks not only learn the mapping from input image to output image, but also learn a loss function to train this mapping. This makes it possible to apply the same generic approach to problems that traditionally would require very different loss formulations. We demonstrate that this approach is effective at synthesizing photos from label maps, reconstructing objects from edge maps, and colorizing images, among other tasks.

Intended uses & limitations:

Used for reconstruction of images from edges

How to use

from torchvision.transforms import Compose, Resize, ToTensor, Normalize
from PIL import Image
from torchvision.utils import save_image
import cv2
from huggan.pytorch.pix2pix.modeling_pix2pix import GeneratorUNet

transform = Compose(
    [
        Resize((256, 256), Image.BICUBIC),
        ToTensor(),
        Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
    ]
)
model = GeneratorUNet.from_pretrained('huggan/pix2pix-uavid-15)

def predict_fn(img):
    inp = transform(img).unsqueeze(0)
    out = model(inp)
    save_image(out, 'out.png', normalize=True)
    return 'out.png'

predict_fn(img)

Limitations and bias

Gives unrealistic colors in the image

Training data

edges2shoes

Training procedure

# clone the repository
git clone https://github.com/huggingface/community-events.git
pip install .

# change directory
cd community-events/huggan/pytorch/pix2pix/

# define config
accelerate config

# launch training with required parameters
accelerate launch train.py --checkpoint_interval 1 --dataset arakesh/uavid-15-hq-mixedres --push_to_hub --model_name pix2pix-uavid-15 --batch_size 2 --n_epochs 50 --image_size 1024 --sample_interval 500

Generated Images

Here,

First Image Row: Input Image
Second Image Row: Generated Image
Third Image Row: Target Image

BibTeX entry and citation info

@article{pix2pix2017,
  title={Image-to-Image Translation with Conditional Adversarial Networks},
  author={Isola, Phillip and Zhu, Jun-Yan and Zhou, Tinghui and Efros, Alexei A},
  journal={CVPR},
  year={2017}
}