## 20221002 BMVC Camera-Ready

1. Provide more examples for comparing the one-stage and two-stage models.
* 這個是不是叫我們要 provide 數據的比較?
* 還是希望我們提供更多圖片的比較?好像比較像希望提供數據的部分，因為我們已經有兩組圖片的比較了。
* [] todo
*
~~3. Provide the links to the dataset we gathered for building SSQD~~
* We have provided the citations in the paper's reference part, do we need to provide another links on the camera-ready version
* [v] Done

~~4. Add interpretation of PSNR/SSIM~~
* The peak signalto-noise ratio (PSNR) metric measures the error relative to the ground truth signal, whereas the structural similarity index (SSIM) metric emphasizes the structure of the image. Our method consistently achieves the 1st or 2nd best scores in terms of both PSNR and SSIM across most datasets, showing that our results not only have smaller errors but also look similar to the ground truth. We will add the interpretation of metrics in the revision. (copy from the rebuttal)

6. Include the OCR result of BEDSR-Scan
* [] todo (may not be the expected result)

8. Provide execution time comparison
* execution time comparison 差異沒有很大
* 放在 supp ? Yes
* [] todo

~~9. Conduct ablation studies of the terms in the Deshading Stage~~
* 有點不懂這題，意思是要加上 deshading stage 的 PSNR 和 SSIM 嗎? Table 2 已經有對SSQD的gt 結果比較。
* 沒有這個結果，略過
~~10. Add two related works~~
* [1]Bogdan, Karina OM, et al. "DDocE: Deep Document Enhancement with Multi scale Feature Aggregation and
Pixel-Wise Adjustments." International Conference on Document Analysis and Recognition. Springer, Cham, 2021.
* [2]Dey, Soumyadeep, and Pratik Jawanpuria. "Light-Weight Document Image Cleanup Using Perceptual Loss."
International Conference on Document Analysis and Recognition. Springer, Cham, 2021.
* 有需要比較嗎? X

~~8. Add acknowledgment for the funding~~
* This work was supported by FIH Mobile Limited and Ministry of Science and Technology (MOST) under grants ***
* 要問一下老師

### Minor Modifications

[v] 修正 Figure 4 的圖片都是 downsample
[v] In Figure 1: missing space between title and next sentence.
[] In caption of Figure 2: “Starting with some shadow maps,” the source of the shadow maps could be better
described.
[v] Moving the ablation studies on the model architecture and limitations from supp to the main paper if there is some space, according to the reviewers' comments.

## I'm confused

* rebuttal 之後都變成 accept
* 最後一個 reviewer 沒有 Final Ratting耶?
* 我們還有需要問 FIH 關於 source code / dataset 的 release 嗎? 如果有的話，子魁之前提醒我要發一封正式的信給他，他再轉寄給法務。
* 這個流程之後，BMVC 應該會有需要拍影片和做poster，還有報告的時候可能還是會有被提問的環節?(我不確定)，這部分可能 dataset 的 release 會再次被詢問。