Compression for EM data

[amotta]

Compression of segmentation data improves both effective read rate and disk space. It might thus be worth to think about compression of the EM (or volumetric image data in general, if we also consider light microscopy, etc.), too.

Flight mode, in particular, has higher requirements for speed than for quality. A scheme, where small amounts of data can be quickly read from disk and sent across the interwebs before being decompressed in the browser could be interesting.

Questions, which might guide us:

• What compression rates are realistically achievable?
• What compression rate do we consider "worth it"? (Possible reference point for comparison: Dataset with 4-bit per voxel).
• Do we allow lossy compression? (Probably good enough for flight mode. But good enough in general?)
• What are the volumetric image compression methods used in other biomedical fields? (JP3D?)
• What about considering a (32 vx)³ bucket as (32 vx)² image with 32 channels and then applying traditional image compression methods?
• What about considering a (32 vx)³ bucket as (32 vx)² video with 32 frames and then applying video compression?