Explore projects

Experimental method to research "implicit optical perspective" in deep learning computer vision systems.

The procedural tools I use in my deanimation artworks.

Uses Python, and depends upon:

FFMPEG for video frame extraction and encoding. Pillow for image manipulation and lighten/darken blending.

Simple scripts that create ART compatible .xmp files from Raw Therapee (.pp3) and Lightroom (.xmp) metadata files. Currently only grabbing the rating value as the colour, crop, sharpening values are vastly different.

This script digitally recreates a night-photography technique that requires long exposure times to average out the reflections off a water body, creating the illusion of a closer-to-flat surface.

Needless to say, the photos should be well aligned for better results. The samples were taken using a cheap tripod for mobile phones, so nothing too fancy.

Phone camera users may have to overcome two different issues while trying to do the long-exposure technique. Some cameras flat out do not allow for very long exposures: a midranger I have does not go beyond 1/12 sec, and another older midranger goes up to 0.4 sec. A lower-midranger I also have, however, happens to allow up to 30 sec. It is a tossup from phone to phone, as this is not the typical usage our devices are tuned for. This lack-of-long exposure is what the present script tackles by averaging a burst of regular short-exposure pictures.

The second issue is that while cameras may have the ability to tune the light intake, via the diaphragm, phone cameras typically cannot. The long exposure needed for the averaging-out to take place can result in a burned image parts. This can be mitigated by using a Neutral Density (ND) filter, some of which come as part of phone-camera addon sets. Another alternative is to directly work with short exposures, take a burst of pictures and apply the averaging process implemented in the present script.

For comparison purposes, find in the TrueLongExposure folder an image made with the long exposure method and the ND32 filter.

Regular cellphone photography often produces high resolution images, i.e., images rendered into 12MP or 16MP are practically the norm. However, a number of these resulting images may not truly contain full 12MP or 16MP of information due to low light conditions, shaky hands or other blur-inducing conditions. As someone who can return from a trip with 10GB of photos, a lot of times I do not need the 12MP or 16MP resolution, or the pictures themselves may not warrant their pixel count, as stated above. Therefore, for storage reasons I often apply a downsampling factor of 2 and sometimes 3, leading to a fourfold or ninefold reduction in drive usage. As the number of image files can go up to hundreds or thousands, I scripted the downsampling procedure. Nothing less, nothing more. I made two versions. There is a python version that reduces the file in place, i.e., replaces the original files with their downsampled versions, and as such reduction is irreversible, the code produces a file, which prevents successive downsampling. The .sh version outputs the downsampled files into the out/ directory, and is safer.

Entangle camera control and capture

My photography portfolio, which is avalaible at https://snap-it.boreux.tech

Import Flickr data into a Pixelfed account