PaperEdge

The code and the DIW dataset for "Learning From Documents in the Wild to Improve Document Unwarping" (SIGGRAPH 2022)

[paper] [supplementary material]

Documents In the Wild (DIW) dataset (2.13GB)

link

Pretrained models (139.7MB each)

Enet

Tnet

DocUNet benchmark results

docunet_benchmark_paperedge.zip

The last row of adres.txt is the evaluation results. The values in the last 3 columns are AD, MS-SSIM, and LD.

Infer one image.

1. Download the pretrained model to the models directory.
2. Run the demo.py by the following code:

   $ python demo.py --Enet_ckpt 'models/G_w_checkpoint_13820.pt' \
                    --Tnet_ckpt 'models/L_w_checkpoint_27640.pt' \
                    --img_path 'images/1.jpg' \
                    --out_dir 'output'
   

3. The final result: