NJUVISION / rho-vision

ρ-Vision: Efficient Visual Computing with Camera RAW Snapshots
https://njuvision.github.io/rho-vision
MIT License
39 stars 3 forks source link

Initialization of wb and ccm parameters in ISPpipeline & Issue with reading RAW images #12

Open XiaMu799 opened 4 months ago

XiaMu799 commented 4 months ago

Hello,

I have two questions regarding the ISPpipeline implementation:

Initialization of wb and ccm Parameters: Could you please explain how the wb (white balance) and ccm (color correction matrix) parameters are initialized in the ISPpipeline? Are there any specific methods or default values used for this initialization?

RAW Image Reading Issue: I am facing an issue with reading RAW images using the provided code. It seems that the RAW images cannot be read properly. Could you kindly re-share the RAW images or provide guidance on how to correctly read them?

Thank you for your assistance!

lizhihao6 commented 4 months ago

Hello,Thank you for your inquiry.

1.We initialize the white balance (wb) and color correction matrix (ccm) parameters using the camera's built-in default values as we mentioned in the paper. These can be accessed through rawpy.

2.Could you specify which RAW image you are having trouble reading?

XiaMu799 commented 4 months ago

Hello, it is a great honor to receive your letter, I downloaded the LUCID_TRI054S folder using your link on github. When reading a raw image using your read method, the error is as follows: "raise ValueError("Unsupported file format.") ; ValueError: Unsupported file format. Then I used the software RawTherapee that read raw to not recognize the raw image. I'm sorry to bother you, but I'd like to ask you three more questions. 1. I see that you use the architecture of "Model-based image signal processors via" in this paper. Do you have the code? I can't open the code URL in his github, if there is, can you share it? 2. I see that your data set is unpaired images. Is PSNR not calculated in the index of image generation? Just calculated FID and IS? 3. Are the wb and ccm parameters read from rawpy the average value of the image? I see in the code that wb and ccm parameters do not enter the network learning parameters.

 

なつめ @.***

 

------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2024年7月14日(星期天) 下午2:07 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [NJUVISION/rho-vision] Initialization of wb and ccm parameters in ISPpipeline & Issue with reading RAW images (Issue #12)

您好,感谢您的询问。

1.我们使用论文中提到的相机内置默认值初始化白平衡 (wb) 和色彩校正矩阵 (ccm) 参数。这些可以通过 rawpy 访问。

2.您能否具体说明一下您无法读取的是哪种 RAW 图像?

— 直接回复此电子邮件、在 GitHub 上查看或取消订阅。 您之所以收到此邮件,是因为您是该主题的作者。消息ID :< @.*** >

lizhihao6 commented 4 months ago

Hi, なつめ

  1. If you are referring to the paper titled "Model-Based Image Signal Processors via Learnable Dictionaries," I do not have the code either. I re-implemented it myself.

  2. I currently do not have the Lucid dataset. If you can provide a demo image, I can help you read it.

XiaMu799 commented 4 months ago

Hi, なつめ

  1. If you are referring to the paper titled "Model-Based Image Signal Processors via Learnable Dictionaries," I do not have the code either. I re-implemented it myself.
  2. I currently do not have the Lucid dataset. If you can provide a demo image, I can help you read it.

Thank you very much. Attached is a raw image of lucid. I see that there are iphone_xsmax and huawei_p30pro in the data set folder multiRAW. How can I download these data sets? [Uploading LUCID_TRI054S-C_223800455__20231105132456508_image0.zip…]()

lizhihao6 commented 4 months ago

The URL you provided is not accessible. The other datasets, including the iPhone XS Max and Huawei P30 Pro, can be downloaded using the links provided in the README file as https://box.nju.edu.cn/d/0f4b5206cf734bd889aa/.

XiaMu799 commented 4 months ago

您提供的 URL 无法访问。其他数据集(包括 iPhone XS Max 和 Huawei P30 Pro)可以使用 README 文件中提供的链接下载,网址为https://box.nju.edu.cn/d/0f4b5206cf734bd889aa/。

Thank you very much for sharing the data set. I trained invISP with rgb images from BDD100K and raw images from iPhone XSmax. The object detection of RAW image I understand is to generate simRAWb through the rgb of bdd100k, use simRAWb to train the object detection of RAW, and test the raw and rgb pairs of images in the iPhone for testing, right? But I did not find the detection model in github, could you please share it?