Class Discussion Summary (Jan 19)

Scribes:

Anthony Liu, Yishu Wang, Alexander Cardaras, Alex Salman, Niharika Srivastav, Ishaan Paranjape

Put the citations to papers discussed here:

• Coded aperture (builds on deconvolution approach) (introduces coded aperture) (has code)
  • Anat Levin, Rob Fergus, Frédo Durand, and William T. Freeman. 2007. Image and depth from a conventional camera with a coded aperture. In ACM SIGGRAPH 2007 (Links to an external site.)
• Night-sight (builds on burst imaging approach) (example modern feature)
  • Orly Liba, Kiran Murthy, Yun-Ta Tsai, Tim Brooks, Tianfan Xue, Nikhil Karnad, Qiurui He, Jonathan T. Barron, Dillon Sharlet, Ryan Geiss, Samuel W. Hasinoff, Yael Pritch, Marc Levoy, “Handheld mobile photography in very low light.” ACM Transactions on Graphics (TOG) 38, no. 6 (2019): 1-16. (Links to an external site.)
  • https://ai.googleblog.com/2018/11/night-sight-seeing-in-dark-on-pixel.html Links to an external site.

Write the Class Discussion Summary Here:

• The first paper describes a de-convolution method. But we saw that before. What are the key-contributions that allowed this paper to be accepted? There may be more than one, or there may be one primary contribution more important than the rest.

The novel idea present in this paper is that the coded aperture allowed for the retention of both RGB and depth information from a single image frame. As a side effect, the coded aperture also removed ringing present in a conventional aperture. The authors were also able to calculate the depths using refocusing. Here different blur scales, which correlate to the shape of the blur, give us depth information.

• The second paper describes a burst imaging algorithm. But we saw that before. What are the key-contributions that allowed this paper to be accepted? There may be more than one, or there may be one primary contribution more important than the rest.

Motion metering uses prediction that dynamically calculates exposure time based on the static and dynamic objects in the scene. Applying burst image algorithm to get auto white balancing in low light photography.

• The coded-aperture idea has not seen wide-spread adoption, while the low light imaging paper describes a commercial product. What do you think is preventing the coded-aperture idea from appearing in consumer products?

Coded aperture disables the control of the lighting. In detail, it fails to control the amount of light entering, thus less flexible. Coded apertures that are too small may cause diffractions as well. This framework also restricts the focus depth to a range of 2-3m. Also, in industrial scenarios, opening the lens body to switch aperture is either extremely time-consuming or dust-proof, so not ideal in terms of efficiency and sustainability. Finally, the amount of time to process images is way too large.