The Lumix S1R II's 177 MP frame, checked by a second decoder: 176.9 million sensels, none different

LibRaw's November snapshot opens the S1R II's tiled 177 MP RW2 that its release refused. Every sensel matches ShootFlow's reader.

This morning ShootFlow learned to open the Panasonic Lumix S1R II's high-resolution frames (the earlier post). These are 16296 x 10856 sensels stored as a 4 x 2 grid of compressed tiles. The one public sample is P1000450.RW2 on raw.pixls.us. At that point nothing else could open it. LibRaw 0.22.2, which every earlier RW2 check was compared against, refused it. So the decode was judged by indirect checks: smooth seams at the tile edges, our C and Python decoders agreeing, and the embedded preview. Those checks are strong evidence but not proof.

A second decoder

LibRaw's git master has a newer RawFormat 8 path than its releases (snapshot 202611, dated 4 November). It builds from source in about 30 seconds. A small helper, tools/libraw_dump.cpp, writes out its raw sensels and metadata, and tools/libraw_dump_check.py compares them with ours, field by field. It runs the same comparison the rawpy-based check has made on every other raw format.

The result: all 176,909,376 sensels match. So do the as-shot white balance (R/G 1.836, B/G 2.105), the black level (512), the CFA (RGGB), the in-camera crop (16288 x 10848 at 4, 4) and the white level (16319). Two files the old release already confirmed match the snapshot too: a normal S1R II frame and an S5 II 96 MP frame. With one sensel flipped in the dump, the check fails at that sensel. (Full write-up.)

Why the release refused it, and where the two decoders differ

LibRaw 0.22.x models RawFormat 8 as at most five vertical stripes, each the full height of the sensor. A second row of tiles cannot pass that test. The snapshot replaces the stripes with a real tile grid. It finds each tile's top by adding up the heights of the tiles above it. It never reads the top from the file: the tag holds (top << 16) | left, and LibRaw stores it in a 16-bit field, which keeps only the left edge.

ShootFlow reads both halves and refuses a tile whose stored top disagrees with the sum of the heights above it. On every real file the two numbers agree, so both decoders produce the same sensels. On a damaged file they would part ways: LibRaw would place the tile by its heights, and ShootFlow would stop and say why.

What is still open

Adobe DNG Converter on the same frame would make a third reader, and it needs a machine that has it. Colour for the S1R II still needs its camera profile, as for every RW2.

All writing · More in Tools and projects · More on ShootFlow · All tools and research