-
One BLAS call made our raw developer's last bits depend on the CPU
Moving ShootFlow's noise-aware blend to C meant replacing a numpy matrix product whose fused multiply-adds varied by machine. Same bytes, 43% less added time.
-
Every Phase One raw carries its own flat field: applying it matches LibRaw to 3 codes on ten backs
IIQ files hold a factory calibration: flat fields, defect maps, tile curves. ShootFlow now applies it; outside dead columns it matches LibRaw within 3 codes.
-
An aerial camera's 5% blue gap was not the calibration: LibRaw skips Phase One's offset tables on IIQ S
Phase One calibration checked on 13 more files, 1.4 billion sensels. The iXU180's colour gap came from black offset tables LibRaw never reads on IIQ S.
-
Taking BLAS out of a raw developer's noise read: same bytes on 12 cameras, and no longer machine-dependent
ShootFlow's patch-PCA noise estimate now sums in a fixed pairwise order in C and numpy alike. 12 of 12 raws develop to the same bytes; the step is 1.25x faster.
-
Same bytes, a third less added time: moving a raw developer's noise reading to C
ShootFlow's noise-aware demosaic spent 1.4 s per 24 MP frame reading noise. In C it takes 0.4 s, and 14 of 14 sample raws still develop to the same bytes.
-
Summing in numpy's order: a C kernel that keeps ShootFlow's raw developer bit-exact
ShootFlow's noise-aware demosaic got a C block-statistics kernel that reproduces numpy's pairwise summation: same bytes on 12 real raws, 1.11x faster.
-
A 2004 Kodak raw file, decoded sensel for sensel
ShootFlow now reads Kodak DCS Pro DCR files. On the one public DCR, all 13.6 million sensor values matched LibRaw's, and the C decoder reads it in 0.13 s.
-
Nine real Canon R5 files: every sensel right, and a black level 41 levels low
ShootFlow's CR3 reader matched LibRaw on 290 M sensels of real R5, R5 Mark II and R3 files, but the R5 Mark II's black and white balance were wrong. Now fixed.
-
Burst grouping for a 10,000-photo cull: the same groups in 5.8 s instead of 115
ShootFlow's cull now finds burst groups in C on every core: identical groups to the Python index, 10,000 frames in 5.8 s instead of 115 s.
-
Burst grouping that survives a wide pan: 10% pans went from 7-28% grouped to 86-95%
ShootFlow's cull compared only one frame's centre window, so pans past 6% split bursts. Off-centre window pairs fix it: held-out pair recall 0.91 to 0.985.
-
Burst grouping that survives a zoom: 10% zoomed frames went from 40% grouped to 99%
ShootFlow's cull now checks each frame against smaller windows of its neighbour, so frames zoomed 3-10% tighter stay in one burst. Held-out: 0.72 to 0.91.
-
Canon's tiled CR3s were on the older cameras: 39 raw files, 21 bodies, checked against LibRaw
We guessed the R5 wrote tiled CR3s. It didn't. The EOS R, RP, M50, 90D and R100 do, and our reader refused all of them until today.
-
Twelve Coolpix raw files read as noise, and nothing complained
ShootFlow's NEF reader took Nikon Coolpix NRW files without an error and decoded them into noise. Five storage layouts later, 13 of 13 match LibRaw.
-
Minolta's last raw files, from the DiMAGE 7 to the Dynax 7D, read sensel for sensel
ShootFlow now reads Minolta and Konica Minolta MRW raw files directly. On nine bodies every sensor value matched LibRaw 0.22.
-
Phase One IIQ files, from the P25 to the IQ3 100MP, developed with the colour matrix they carry
ShootFlow now reads Phase One IIQ L raw files directly and needs no camera profile. On ten backs every sensor value matched LibRaw 0.22.
-
Grouping 10,000 photos into bursts without comparing every pair
ShootFlow's cull compared every pair of photos to find bursts. A pigeonhole index on hash chunks gives the same groups 6x faster at 10,000 frames.
-
Phase One IIQ S, read sensel for sensel: 547 million values equal to LibRaw
ShootFlow now reads Phase One's smaller IIQ S raws, the setting aerial iXU and iXM cameras shoot. Six real files, 547 M sensels, none different from LibRaw.
-
Phase One IQ4 16-bit IIQ, read directly, and two colour matrices that disagree
ShootFlow now reads 16-bit IIQ L from the IQ4 150MP and Trichromatic: 520 M sensels equal to LibRaw. The IQ4's own matrix and LibRaw's disagree on colour.
-
Phocus .fff files scale red and blue but keep the old white balance
ShootFlow now develops Hasselblad Phocus .fff files. Newer exports scale red and blue but keep the old neutral, and the data shows which files did it.
-
The cull split 7% and 9% pans because nobody tested them: windows at every whole percent
ShootFlow's burst grouping matched windows on a 3% grid, so pans that moved content between steps split. 24 more windows: held-out recall 0.982 to 0.998.
-
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.
-
The card noise floor, closed: on 24 unseen cameras the true noise model loses too
A second-lowest card floor halved the cost of the minimum but still missed its bar, and the rows it was meant to fix lose even with the exact noise.
-
A card-level noise floor on 182 real cameras: three rules committed first, none passed
ShootFlow reads a raw file's noise from the frame. Capping that reading at the card's lowest fixed 17 of 24 losses but missed its pre-set bar by 0.005 dB.
-
Hasselblad 3FR without Phocus: ten bodies match LibRaw, and the H4D-60's shadows were wrapping
ShootFlow now develops Hasselblad 3FR files directly, with no profile. On ten bodies every sensel matches LibRaw, and one body's noise wrapped past zero.
-
The Lumix S1R II's 177 MP frames: one tag held two numbers, and LibRaw can't open them either
A search for Panasonic RW2 tone curves found none in 35 files. It did find a 4 x 2 tile grid that both our reader and LibRaw refused, and why.
-
The 6% pan was the gate, not the grid: one constant took those pans from 85% grouped to 100%
ShootFlow's cull split 6% pans more often than 8% ones. A finer window grid was the guess. The cause was a whole-frame gate: 24 to 44, no added false merge.
-
Split 10% pans in a photo cull: the wrong window pair, not too coarse a grid
ShootFlow's cull split one in eight 10% pans. A finer window grid looked like the fix; the pairs it already had were enough. Held out: 0.87 to 0.98.
-
Our 16-bit raw masters held 12 bits: one table read, fixed
ShootFlow's sRGB encode read a 4,096-entry table at the nearest entry. Interpolating a 16,384-entry one lifts 16-bit masters from 4,096 levels to 45,000-65,536.
-
Keeping a burst together when the camera moves: crop-tolerant grouping in ShootFlow's cull
A perceptual hash split bursts the camera panned 4% through. Matching 88% windows took held-out burst-pair recall from 0.40 to 0.72 for 2 more false merges.
-
Halving one array in a raw develop, and the 12-bit tone table it turned up
ShootFlow's noise-aware develop holds its bilinear image in float32: 15% less memory, and every changed pixel value is one step of a 4,096-entry tone table.
-
Same bytes, 2.2x faster: banding ShootFlow's noise-aware demosaic
ShootFlow's noise-aware raw develop now runs in row bands on every core: 2.2x faster and 41% less memory on 14 real raws, every output byte-identical.
-
Raws with no noise profile now read their own noise: 52 real files through the new default
ShootFlow's noise-aware demosaic now reads the noise from the frame when a raw has no NoiseProfile. All 52 sample raws took it: what it changed and cost.
-
Why the noise scale stays at 0.8: one held-out tile decides it
We tried 861 rules to push ShootFlow's frame-read noise scale above 0.8, all picked on tune tiles only. Every one that won on tune lost on one held-out crop.
-
Pushing a 16-bit raw master 3 stops: what the old sRGB table cost
Lifted 3 stops, ShootFlow's old 16-bit masters skipped up to 39 of 256 screen codes in the shadows. The interpolated encode skips none on five real camera DNGs.
-
A card copy that only took the raws it could develop left 6 of 71 behind
ShootFlow's card copy skipped raw formats it could not develop. It now takes 27, lists what it leaves, and develops Leica RWL files.
-
Canon C-RAW in tiles: read, checked against LibRaw, and a LibRaw seam bug found on the way
ShootFlow now reads C-RAW CR3 files stored in tiles. Side-by-side tiles match LibRaw exactly; stacked tiles exposed a misplaced test in LibRaw's decoder.
-
Making a learned demosaic twice as fast without changing a single output bit
We fused a learned Bayer demosaic's 30 feature passes into one row pipeline. It runs 1.8-2.2x faster and matches the numpy reference bit for bit on seven raws.
-
A 2003 Canon raw file, read sensel for sensel without converting it first
ShootFlow now reads Canon CRW files from 15 bodies of 2000-2004 directly. On 20 real files every sensor value matched LibRaw 0.22.
-
Finding closed eyes on real photos: a face-landmark model against the old Haar cascades
On 105 real CC0 photos, a MediaPipe landmark detector flagged 5 of 6 closed-eye shots and no false ones. The Haar cascades flagged 3 of 6, with 8 false ones.
-
The demosaic weight should follow colour correlation, we said. Real photographs disagreed
A pre-registered test of a colour-correlation feature for ShootFlow's learned demosaic weight. It missed its bar, and real photographs ran the other way.
-
Closed eyes on a face cut in half by the frame: mirror the edge
A face cut by the frame edge is invisible to MediaPipe's face finder. Mirroring the edge strip finds 20 of 40 half faces, 19 read the same as the whole face.
-
The learned Bayer demosaic in C: bit-identical to numpy, and why it is only 2x faster
ShootFlow's learned demosaic weight now runs in C, equal to numpy bit for bit on seven real raw frames. It is 2-3x faster, not 4-5x: the features cost the time.
-
Teaching a demosaic weight about noise worked a little; teaching it fine detail backfired
ShootFlow's learned demosaic weight, refit on noisy mosaics, gains a little under noise. Synthetic fine detail made it worse, for a reason about colour.
-
Testing an X-Trans demosaic on scenes it never saw: borrowed Bayer frames as truth
Only two X-Trans raws are public, so we mosaicked six other cameras' scenes with Fujifilm's layout. We led LibRaw's three passes under noise on all seven.
-
Most cameras were still getting the softest demosaic. We replaced it and scored it against LibRaw's seven
A directional Bayer demosaic for ShootFlow beats bilinear on every test and has the best colour of LibRaw's seven methods on five of six. Where it loses, too.
-
A 1997 demosaic formula cost our sharpest photos a decibel of detail. We changed one number
ShootFlow's Bayer demosaic lost luma to plain bilinear on fine, aliased texture. A smaller Laplacian weight closes most of it and lifts colour on every test.
-
Canon C-RAW decoded: a 5/3 wavelet, a QP grid and every sensel equal to LibRaw's
ShootFlow now reads Canon C-RAW directly: on three real files from an EOS R8 and an EOS R3, every sensor value equals LibRaw's, in under a second.
-
Reading Canon CR3 raw files without a converter, sensel for sensel
ShootFlow now develops full-quality Canon CR3 raw files directly: on two real EOS R8 frames, all 25.5 million sensor values match LibRaw's.
-
Reading Canon's tiled CR3 without a tiled file: we made some and asked LibRaw
ShootFlow now reads full-quality CR3 files stored in tiles. No public tiled file was reachable, so real R8 frames were re-tiled and LibRaw read them back.
-
Every current Lumix raw file now develops without a DNG converter
ShootFlow now reads RawFormat 8, the RW2 layout of every Lumix since 2022. On ten real files from six bodies, all 464.6 million sensor values equal LibRaw's.