A unified framework for event-based video. Encoder/transcoder/decoder for ADΔER (Address, Decimation, Δt Event Representation) video streams.
Bot releases are hidden (Show)
Support Prophesee data files (.dat extension)
Much-improved transcoding of pure DVS-style data. Will be replicated for .aedat files at some point (so they won't have rely on the davis-edi-rs dependency
For best results with Prophesee files in adder-viz:
-- Set Δt_ref slider to 1
-- Use Auto quality, with slider at CRF 9 (lowest sensitivity to temporal intensity change)
Using the source-modeled ADDER compression, the Prophesee file I've tested gets compressed by about 2:1 over the raw input. For comparison, Gzip makes the file 5x larger. The video had a stationary camera, though, which greatly benefits the current ADDER compression model since it doesn't have motion compensation. Your mileage may vary.
Also note that the compression/decompression speeds here are pretty slow, due to the very high event rate. Compressing 1 second of data may take several seconds in actuality. Putting the codec on a separate thread group will help, but further optimization is necessary to make it real time for DVS-based sources.
The feature detection application works as it does with other source video types. Just check the Detect features box in adder-viz, and you can visualize them best with the Show instant option. Feature detection increases the sensitivity of nearby pixels, making the output file size larger. I'll eventually add an option to disable feature-driven sensitivity adjustment.
Special thanks to @arghasen10 for his help and for providing a Prophesee test file.
Published by ac-freeman 10 months ago
Published by ac-freeman 11 months ago
Published by ac-freeman about 1 year ago
Published by ac-freeman about 1 year ago
Published by ac-freeman over 1 year ago
Transcode from a live DAVIS camera feed
Published by ac-freeman almost 2 years ago
Add support for transcoding DAVIS 346 camera data to ADΔER
Published by ac-freeman about 2 years ago
Use a new graph structure for event pixels which is more amenable to various input types
Published by ac-freeman about 2 years ago
Published by ac-freeman about 2 years ago
Added several example programs
Published by ac-freeman about 2 years ago