Narrate HN: 3D Reconstructions of scenes from the wrestle in Ukraine

Fuck Putin, Слава Україні! 🇺🇦 ✊ 🌻

Other folks who place not search for history are doomed to repeat it, and we can most efficient search for what we file. I’ve been experimenting with creating 3D reconstructions of scenes from Russia’s unprovoked invasion of Ukraine the employ of videos from twitter/telegram/tiktok and youtube. I am making these accessible right here as CC BY-SA, please DM me on twitter or electronic mail me thot@thiic.cc whilst you happen to hang any questions or hang footage to employ for extra reconstructions. Supply code is accessible on github

Reconstructed Scenes

These can also take a moment to load up, they are 10-20MB every. VR viewing needs to be accessible as successfully with everything scaled honest to existence. If anybody has details about the conventional sources of these videos or higher quality variations please contact me.

Plan

The person steps right here were covered barely successfully in masses of areas on the score nonetheless right here are some notes to make it simpler whilst you happen to rob to ought to strive something similar. I could are trying and maintain this up to this point with the leisure else I behold.

1.Spoil up video to photography
ffmpeg makes this straightforward

ffmpeg -i crashed_hind.mp4 -ss 7.5 -to 30 -r 5/1 h%03d.png

-ss 7.5 -to 30 manner job from 7.5 sec to 30 sec
-r 5/1 sets the framerate to 5fps
h%03d.png write to *.png sequence love h001.png, h002.png, . . .

For some videos you ought to eradicate determined sections or job a chunk with higher tempo circulate on the next temporal resolution and it’s factual to undergo in mind of that as a attainable solve to lost tracking.

2. Reconstruct scene

I employ meshroom for this (factual overview of meshroom utilization), On the total I follow the default processing graph with with a pair exceptions. As a second-to-final step I could in most cases add a MeshDecimate node to tame the mesh density a little since the noisy files from compressed cell-phone videos cannot with out a doubt hope to signify that unparalleled ingredient anyway. On the final step I employ LSCM for UV generation in expose to be determined I the truth is hang correct one texture atlas for the final mesh. Sooner or later as a final resort when issues ought to not working I could flip on producing and matching in conserving with the AKAZE feature descriptors to boot to SIFT, in my experience right here is mainly a shot at nighttime and seldom helps tangibly.

I am animated if industrial offerings would non-public a bigger job of this so if anybody has a spare license/credit for any of those please acquire in contact.

3. Easy and prepare mesh

When I’ve purchased a mesh I employ blender are trying and scale the scene as suited I’m in a position to to employ accurate world objects the employ of a combination of publicly accessible automobile blueprints and guesswork. If the mesh wants cleansing up I could additionally non-public that. As soon as issues see factual I could export everything as a *.glb which is in a space to be straight imported into three.js and in overall appears to be like love a decently successfully supported interchange format. The viewer is correct a few strains of JS and barely unparalleled all the heavy lifting is achieved by three.