2022-04-18 09:44:16 +08:00
|
|
|
# PaperEdge
|
2022-10-26 12:18:36 +08:00
|
|
|
<a href="https://huggingface.co/spaces/SWHL/PaperEdgeDemo"><img src="https://img.shields.io/badge/%F0%9F%A4%97-Open%20in%20Spaces-blue"></a><br/>
|
2022-04-18 09:44:16 +08:00
|
|
|
The code and the DIW dataset for "Learning From Documents in the Wild to Improve Document Unwarping" (SIGGRAPH 2022)
|
2022-07-07 11:43:11 +08:00
|
|
|
|
2022-07-25 11:59:25 +08:00
|
|
|
[[paper](https://drive.google.com/file/d/1z_8YaCc3aGWTHqFP55vgpSaEBz_oaQQe/view?usp=sharing)]
|
|
|
|
|
[[supplementary material](https://drive.google.com/file/d/1szMa0D90E9caKVwonnZFduTpla52sKEz/view?usp=sharing)]
|
2022-07-07 11:51:22 +08:00
|
|
|
|
|
|
|
|
|
2022-07-25 11:59:25 +08:00
|
|
|
|
2022-07-07 11:43:11 +08:00
|
|
|
## Documents In the Wild (DIW) dataset (2.13GB)
|
|
|
|
|
[link](https://drive.google.com/file/d/1qAmLurt6bK0ro8PnRz6rBgVs1rfrsdKi/view?usp=sharing)
|
|
|
|
|
|
|
|
|
|
## Pretrained models (139.7MB each)
|
|
|
|
|
|
|
|
|
|
[Enet](https://drive.google.com/file/d/1OVHETBHQ5u-1tnci3qd7OcAjas4v1xnl/view?usp=sharing)
|
|
|
|
|
|
|
|
|
|
[Tnet](https://drive.google.com/file/d/1gEp4ecmdvKds2nzk9CaZb_pLvhRoyAsv/view?usp=sharing)
|
|
|
|
|
|
|
|
|
|
## DocUNet benchmark results
|
|
|
|
|
[docunet_benchmark_paperedge.zip](https://drive.google.com/file/d/1QM3Y5Ty96ydVCQPNqR0_bnMG9oqIQkGm/view?usp=sharing)
|
|
|
|
|
|
|
|
|
|
The last row of `adres.txt` is the evaluation results.
|
2022-07-07 11:51:22 +08:00
|
|
|
The values in the last 3 columns are `AD`, `MS-SSIM`, and `LD`.
|
2022-10-23 13:56:27 +08:00
|
|
|
|
|
|
|
|
## Infer one image.
|
|
|
|
|
1. Download the pretrained model to the `models` directory.
|
|
|
|
|
2. Run the `demo.py` by the following code:
|
|
|
|
|
```shell
|
2022-12-20 13:04:24 +08:00
|
|
|
$ 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'
|
2022-10-23 13:56:27 +08:00
|
|
|
```
|
|
|
|
|
3. The final result:
|
|
|
|
|
|
