nishavarghese15/druva

Dataset proposed in "Self-supervised Monocular Underwater Depth Recovery, Image Restoration, and a Real-sea Video Dataset", ICCV 2023

DRUVA: Dataset of Real-world Underwater Videos of Artifacts

Dataset proposed in "Self-supervised Monocular Underwater Depth Recovery, Image Restoration, and a Real-sea Video Dataset", IEEE International Conference on Computer Vision (ICCV), Paris, France, pp. 12248-12258, October 2023.

Summary of DRUVA

1. DRUVA contains video sequences of 20 different artifacts (artifacts are mainly submerged rocks of dimension 0.5 meter to 1 meter) in shallow waters.
2. There are 20 videos in all, one for each artifact. The videos are approximately 1 minute long and contain an almost 360◦ azimuthal view of the artifacts at a depth of 0.5 meter to 4 meters from the camera.
3. DRUVA is captured using GoPro Hero 10 Black camera with 30 fps and 1920 × 1080 resolution. Camera intrinsics are provided.
4. DRUVA can be used by underwater (UW) researchers for 3D reconstruction, novel view-synthesis using neural radiance fields (NeRFs), video interpolation, and extrapolation, to name a few.

How to access the Dataset?

1. Read the DRUVA release agreement: https://drive.google.com/file/d/1o_5mlieWYT9FwOlj1oDqbhzm6Q-13RAS/view?usp=sharing
2. Obtain a signed copy of the above agreement form.
3. Enter your details in the request form https://docs.google.com/forms/d/e/1FAIpQLSfogZJEnT0VJL-xc8AaA4UXLm4HDZtj5v1Z1OLLrQW25G5ETQ/viewform?usp=sharing
4. Upload the agreement as part of the request form.

After you submit the request form, you will receive an e-mail notification (within 5 working days) with the link to download the dataset.

Acknowledgement

Support provided by the Department of Science and Technology, India through project No. EE1920271DSTX005001 is gratefully acknowledged. We thank Dr. Sundaresh and his team from National Institute of Oceanography, Goa for helping us in collecting the underwater data.

Cite Us

If you use DRUVA, please cite the following work:

@InProceedings{Varghese_2023_ICCV,
    author    = {Varghese, Nisha and Kumar, Ashish and Rajagopalan, A. N.},
    title     = {Self-supervised Monocular Underwater Depth Recovery, Image Restoration, and a Real-sea Video Dataset},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2023},
    pages     = {12248-12258}
}

druva's Issues

Beta-Net

Hi, how do you ensure that the beta estimated by Beta-Net is accurate, thanks!

Question about proj() transformation function

Hi, thank you for your great work again !

Could you help me explain further about the fundamental theory behind the equation (3) transformation function?

I have limited knowledge about two-view geometry, but I think it might be based on the "plane induced homography" from equation (13.2) in Hartley and Zisserman's "Multiple view geometry":

However, I am not certain about how the relative camera pose T multiplying with estimated depth map D in your equation (3) can be derived from the homography equation (13.2).

Could you help me understand more details about the math (or in matrix representation / code implementation) behind this part?

Thank you very much !

Question about code release

Hi, thank you for your great work !

Is it possible that your team could release the source code of USe-ReDI-Net any time soon?

Thank you.