3 ms·
One thing I've always wanted was an AI algorithm for extracting additional detail from multiple RAW camera frames of the same scene. Many photographers will typ
by jiggawatts 6y ago
One thing I've always wanted was an AI algorithm for extracting additional detail from multiple RAW camera frames of the same scene. Many photographers will typically take 10-100 shots of a subject to ensure that at least one picture is a "keeper".
Keeping one frame and discarding the rest is a bit wasteful in a sense. The other frames have useful information that could be extracted by a well-trained AI to provide super-resolution, increased DoF, additional blur or shake reduction, etc...
I've deliberately kept all of my RAW frames, even the not-so-sharp or slightly shaky ones, because I foresee that at some point in the future this will be an automatic thing that tools like Adobe Lightroom will do that maximise the available image quality.
Storage is cheap, but I can never go back in time and photograph my memorable occasions with a better camera from the future...
- erinaceousjones 6y agoVaguely relatable, there is/was stuff like Photosynth [1] which (as far as I recall, from researching this stuff as a CS undergrad) was making use of Bundle Adjustment [2] and detecting landmarks in images using Scale Invariant Feature Transform (SIFT) [3], to estimate camera poses of photos relative to each other in a 3D scene -- even from completely different cameras. By projecting the discarded frames onto your keeper frame, you've potentially got multiple samples of the same pixels and their neigbours... (Although I guess the "every image is a plane" 3D transform is a bit simple and doesn't account for lens distortion) [1] https://en.wikipedia.org/wiki/Photosynth https://en.wikipedia.org/wiki/Photosynth [2] https://en.wikipedia.org/wiki/Bundle_adjustment https://en.wikipedia.org/wiki/Bundle_adjustment [3] https://en.wikipedia.org/wiki/Scale-invariant_feature_transform https://en.wikipedia.org/wiki/Scale-invariant_feature_transf...