Post

Replies

Boosts

Views

Activity

Reply to Radiometric interpretation of Apple ProRAW and Bayer RAW access via AVFoundation
On the radiometric side, the two paths return different data, not two names for the same data. From published readouts of Halide captures on an iPhone 12 Pro main camera (iOS 14.3 era): the Bayer DNG is mosaiced — Color Filter Array, 1 sample per pixel — with BlackLevel 528 and WhiteLevel 4095. That is a 12-bit sensor readout with a real pedestal. The same author's ProRAW file from the same camera reads BlackLevel 0, 0, 0 and WhiteLevel 65535 x3, with NoiseReductionApplied 0.95 against 0 in the Bayer file: normalized linear RGB, already demosaiced and partially processed. A telephoto readout tells the same Bayer story — iPhone 11 Pro telephoto, Halide II, iOS 18.0.1: Color Filter Array, BlackLevel 528. Does true Bayer DNG capture exist via AVFoundation? Yes, through third-party apps — but on Quad Bayer devices it is the binned stream. Reeflex states it directly: its RAW mode produces standard Bayer RAW (CFA) DNGs, and on Quad Bayer iPhones that means 12MP files, because 4 pixels are grouped in a single color in the Bayer pattern. 48MP on those devices exists only as ProRAW. That gap is what FB25352517 asks Apple to close — full-resolution Quad Bayer RAW for third parties — and a supplement answering the DTS engineer's specifics is posted in thread 853978, where he confirmed the feedback is open and under investigation.
1h
Reply to AVFoundation: full-resolution Quad Bayer RAW capture path for third-party computational photography? (FB25352517)
Thank you for confirming FB25352517 is open and under investigation, and for naming the four specifics: devices and cameras, single vs burst, per-frame values, and a sample of the current path. In your order: this supplement has been added to FB25352517 via Add More Information. (a) Devices and cameras. Primary: iPhone 18 Pro Max — and iPhone 18 Pro, with the identical camera system (Apple Tech Specs). All three 48MP Fusion cameras: 1) Fusion Telephoto, 48MP, 100mm (4x), ƒ/2.8 — the core ask; 2) Fusion Main, 48MP, 24mm, variable ƒ/1.48–ƒ/4.0, as the comparison camera; 3) Fusion Ultra Wide, 48MP, 13mm, ƒ/2.2. In API terms: any camera where availableRawPhotoPixelFormatTypes today returns only binned Bayer RAW formats. (b) Bursts. Adobe Research documents Project Indigo merging up to 32 RAW frames per photo, aligned to a reference frame. INFERRED requirement: at least 15 frames, ideally up to 32, at full telephoto resolution, sustained and gap-free at the stream's native cadence (15 = shipped count in Google's published handheld multi-frame SR work; 32 = Indigo's documented ceiling; Adobe publishes no Indigo frame rate). The crop math: at 12x (3x crop of the 4x telephoto), today's 12MP binned stream starts from ~1.33MP per frame, at 16x (4x crop) from ~0.75MP. From a full-res 48MP stream: ~5.3MP and ~3MP. (c) Per-frame values. See the table. Every value must describe the 48MP readout delivered, not the binned mode. One blocker today: exact exposure and ISO are only guaranteed at speed prioritisation; balanced and quality hand exposure to Apple's fusion pipeline. (d) What the current path returns. Plainly: no public Bayer RAW characterization of the iPhone 18 Pro Max telephoto exists. On the closest documented devices, the path returns a single-frame, 12MP-class, mosaiced CFA DNG; 48MP exists only as ProRAW (linear, demosaiced). Reeflex's FAQ states that 12MP Bayer cap (iPhone 14/15 Pro series). Published readouts: iPhone 11 Pro telephoto, Halide II, iOS 18.0.1 (ExifTool) — Color Filter Array, 1 sample per pixel, BlackLevel 528. iPhone 12 Pro main, Halide, iOS 14.3 era — BlackLevel 528, WhiteLevel 4095. The same author's ProRAW file reads BlackLevel 0 and WhiteLevel 65535. Decode path, part of the request: the WWDC26 panel (session 8018) said Quad Bayer DNG output needs CIRAWFilter decode support. The ask is the trio: full-res capture format, a defined DNG representation (4x4 CFA layout and levels), and CIRAWFilter Quad Bayer decode. Every figure above traces to the published sources named in the text. Per-frame valueNeededExposed todayCFA layout4x4 Quad Bayer layout, true for the 48MP modePartial: CFA, 1 sample/pixel, plane colours in files; exact CFAPattern unpublished — UNVERIFIEDBlack levelTrue 48MP-mode value, per frame528 on 11 Pro telephoto and 12 Pro main (see (d)); Quad Bayer value UNVERIFIEDWhite levelTrue 48MP-mode value, per frame4095 on 12 Pro main; Quad Bayer value UNVERIFIEDExposureActual value per frame, delivered with the bufferIn single-frame EXIF; burst stability undocumentedISOActual value per frame, delivered with the bufferIn single-frame EXIF; burst stability undocumentedTimestampsSub-frame-interval, monotonic, per frame, for alignmentWall-clock at 1-second resolution only in published readouts; no alignment-grade timestamp
1h
Reply to Radiometric interpretation of Apple ProRAW and Bayer RAW access via AVFoundation
On the radiometric side, the two paths return different data, not two names for the same data. From published readouts of Halide captures on an iPhone 12 Pro main camera (iOS 14.3 era): the Bayer DNG is mosaiced — Color Filter Array, 1 sample per pixel — with BlackLevel 528 and WhiteLevel 4095. That is a 12-bit sensor readout with a real pedestal. The same author's ProRAW file from the same camera reads BlackLevel 0, 0, 0 and WhiteLevel 65535 x3, with NoiseReductionApplied 0.95 against 0 in the Bayer file: normalized linear RGB, already demosaiced and partially processed. A telephoto readout tells the same Bayer story — iPhone 11 Pro telephoto, Halide II, iOS 18.0.1: Color Filter Array, BlackLevel 528. Does true Bayer DNG capture exist via AVFoundation? Yes, through third-party apps — but on Quad Bayer devices it is the binned stream. Reeflex states it directly: its RAW mode produces standard Bayer RAW (CFA) DNGs, and on Quad Bayer iPhones that means 12MP files, because 4 pixels are grouped in a single color in the Bayer pattern. 48MP on those devices exists only as ProRAW. That gap is what FB25352517 asks Apple to close — full-resolution Quad Bayer RAW for third parties — and a supplement answering the DTS engineer's specifics is posted in thread 853978, where he confirmed the feedback is open and under investigation.
Replies
Boosts
Views
Activity
1h
Reply to AVFoundation: full-resolution Quad Bayer RAW capture path for third-party computational photography? (FB25352517)
Thank you for confirming FB25352517 is open and under investigation, and for naming the four specifics: devices and cameras, single vs burst, per-frame values, and a sample of the current path. In your order: this supplement has been added to FB25352517 via Add More Information. (a) Devices and cameras. Primary: iPhone 18 Pro Max — and iPhone 18 Pro, with the identical camera system (Apple Tech Specs). All three 48MP Fusion cameras: 1) Fusion Telephoto, 48MP, 100mm (4x), ƒ/2.8 — the core ask; 2) Fusion Main, 48MP, 24mm, variable ƒ/1.48–ƒ/4.0, as the comparison camera; 3) Fusion Ultra Wide, 48MP, 13mm, ƒ/2.2. In API terms: any camera where availableRawPhotoPixelFormatTypes today returns only binned Bayer RAW formats. (b) Bursts. Adobe Research documents Project Indigo merging up to 32 RAW frames per photo, aligned to a reference frame. INFERRED requirement: at least 15 frames, ideally up to 32, at full telephoto resolution, sustained and gap-free at the stream's native cadence (15 = shipped count in Google's published handheld multi-frame SR work; 32 = Indigo's documented ceiling; Adobe publishes no Indigo frame rate). The crop math: at 12x (3x crop of the 4x telephoto), today's 12MP binned stream starts from ~1.33MP per frame, at 16x (4x crop) from ~0.75MP. From a full-res 48MP stream: ~5.3MP and ~3MP. (c) Per-frame values. See the table. Every value must describe the 48MP readout delivered, not the binned mode. One blocker today: exact exposure and ISO are only guaranteed at speed prioritisation; balanced and quality hand exposure to Apple's fusion pipeline. (d) What the current path returns. Plainly: no public Bayer RAW characterization of the iPhone 18 Pro Max telephoto exists. On the closest documented devices, the path returns a single-frame, 12MP-class, mosaiced CFA DNG; 48MP exists only as ProRAW (linear, demosaiced). Reeflex's FAQ states that 12MP Bayer cap (iPhone 14/15 Pro series). Published readouts: iPhone 11 Pro telephoto, Halide II, iOS 18.0.1 (ExifTool) — Color Filter Array, 1 sample per pixel, BlackLevel 528. iPhone 12 Pro main, Halide, iOS 14.3 era — BlackLevel 528, WhiteLevel 4095. The same author's ProRAW file reads BlackLevel 0 and WhiteLevel 65535. Decode path, part of the request: the WWDC26 panel (session 8018) said Quad Bayer DNG output needs CIRAWFilter decode support. The ask is the trio: full-res capture format, a defined DNG representation (4x4 CFA layout and levels), and CIRAWFilter Quad Bayer decode. Every figure above traces to the published sources named in the text. Per-frame valueNeededExposed todayCFA layout4x4 Quad Bayer layout, true for the 48MP modePartial: CFA, 1 sample/pixel, plane colours in files; exact CFAPattern unpublished — UNVERIFIEDBlack levelTrue 48MP-mode value, per frame528 on 11 Pro telephoto and 12 Pro main (see (d)); Quad Bayer value UNVERIFIEDWhite levelTrue 48MP-mode value, per frame4095 on 12 Pro main; Quad Bayer value UNVERIFIEDExposureActual value per frame, delivered with the bufferIn single-frame EXIF; burst stability undocumentedISOActual value per frame, delivered with the bufferIn single-frame EXIF; burst stability undocumentedTimestampsSub-frame-interval, monotonic, per frame, for alignmentWall-clock at 1-second resolution only in published readouts; no alignment-grade timestamp
Replies
Boosts
Views
Activity
1h