Fast Lossless Depth Image Compression
summary
Claims a new lossless compressor for 16‑bit single‑channel depth images (Kinect style) called RVL that encodes alternating run‑length variable‑length runs of zeros and nonzeros plus successive nonzero deltas using base‑8 variable lengths and zig‑zag sign mapping. Implementation is simple (<100 lines of C), beats PNG/JPEG‑XR/JPEG‑LS/RLGR on speed while matching compression, often reaching about 4× compression, and in an eight‑camera Kinect v2 setup reduces network load from ~1 Gbps to ~500 Mbps and restores solid 30 Hz.
AI-written summary · about
chapters (9)
- 0:00Introduction and background
- 1:52Problem with bandwidth and latency
- 3:39Why lossless compression matters
- 6:10Introducing RVL technique
- 9:01Evaluation and results
- 10:27Demo and impact on frame rate
- 11:30Latency complexity and future work
- 12:56Code availability and conclusion
- 13:19Q&A
related talks
25mOptimal Video Compression Using Pixel Shift Tracking
16mGRACE: Loss-Resilient Real-Time Video through Neural Codecs
17mEfficient & Frugal Image Data Processing Pipelines
52mThe JPEG is dead! Long live the JPEG!
01h 11mMoving Pictures
22mAI Enhanced GPU Video Coding: Achieving Joint High Compression Efficiency and Throughput
43mInto the Deep – or what would the 80s have done with depth sensing technology?
26mNew video coding techniques under consideration for VP10 - the successor to VP9