splinter21 / mural-image-inpainting

Mural image inpainting, Mural damage repair, Line drawing guided image inpainting

Line Drawing Guided Progressive Inpainting of Mural Damages - MuralNet

This is the source code for 'Line Drawing Guided Progressive Inpainting of Mural Damages'. We provide the code, the dataset, and the pretrained models. For convienence of introduction, we name the proposed network as MuralNet.

Abstract:

Mural image inpainting refers to repairing the damage or missing areas in a mural image to restore the visual appearance. Most existing image-inpainting methods tend to take a target image as the only input and directly repair the damage to generate a visually plausible result. These methods obtain high performance in restoration or completion of some specific objects, e.g., human face, fabric texture, and printed texts, etc., however, are not suitable for repairing murals with varied subjects, especially for murals with large damaged areas. Moreover, due to the discrete colors in paints, mural inpainting may suffer from apparent color bias as compared to natural image inpainting. To this end, in this paper, we propose a line drawing guided progressive mural inpainting method. It divides the inpainting process into two steps: structure reconstruction and color correction, executed by a structure reconstruction network (SRN) and a color correction network (CCN), respectively. In the structure reconstruction, line drawings are used by SRN as a guarantee for large-scale content authenticity and structural stability. In the color correction, CCN operates a local color adjustment for missing pixels which reduces the negative effects of color bias and edge jumping. The proposed approach is evaluated against the current state-of-the-art image inpainting methods. Qualitative and quantitative results demonstrate the superiority of the proposed method in mural image inpainting.

-Mural image inpainting, -Mural damage repair, -Line drawing guided image inpainting

Mural Inpainting with and without line drawings

Network Architecture

Set up:

Requirments

• Python 3.7
• PyTorch 1.6
• We run the codes on NVIDIA GPU RTX2080ti with CUDA 10.2 and cuDNN 7.6.5

Preparation:

Data Preparation

We collected a mural dataset DhMural1714 to train and test the MuralNet. You can download them and put them into "./dataset/".

Pretrained Models

Our model is pretrained on DhMural1714, the pretrained model can be obtained at ./checkpoints/InpaintingModel_gen.pth and ./checkpoints/InpaintingModel_dis.pth

Download:

DhMural1714 Dataset

In our research, we collected some replicas of murals by artists in addition to the real mural paintings. The real mural paintings are captured by digital cameras, while the replicas of murals are obtained by scanning the album. A total of 1,714 images are collected. Among them, there are 525 real murals and 1,189 replicas.

dataset: https://1drv.ms/u/s!AittnGm6vRKLzXorf1nkiDPRQB4D?e=Avv27i

You can also download the dataset from
Link: https://pan.baidu.com/s/13dkh4zX3C4_z2W3Kigg6uQ passcodes: 0vlr

Pretrained Models

You can also download the pretrained models from
link: https://pan.baidu.com/s/1EKvTAHyOaMbL9s7aqdZEfw passcodes: vg5v

Training:

MuralNet is trained on two stages: 1) training the coarse network, 2) training the whole model.

To train the model, modify the training parameters in checkpoints/config.yml.

Run the code for training:

python train.py

Test:

We provide several example images in checkpoints/test for testing, just run the code:

python test.py

Directory of testing images can be modified in main.py, the network requires the input images, the corresponding line drawings, and the masks for inpainting.

Evaluation:

To evaluate the performance, run the code:

python eval_mix.py

Acknowledgment:

We implement our method by referring to EdgeConnect. We thank the authors of the EdgeConnect paper. If you use this code in your work, you should also cite the EdgeConnect paper.

@inproceedings{nazeri2019edgeconnect,
  title={EdgeConnect: Generative Image Inpainting with Adversarial Edge Learning},
  author={Nazeri, Kamyar and Ng, Eric and Joseph, Tony and Qureshi, Faisal and Ebrahimi, Mehran},
  journal={arXiv preprint},
  year={2019},
}