Camera RAW — Bayer Mosaics, Demosaicing & RAW Formats

01 — What RAW is

The term RAW describes a recording format where the camera’s image signal processor (ISP) decisions are deferred to post-production. Every camera applies some processing before writing to storage: black level subtraction, pixel defect correction, analog gain. No camera captures pure, unprocessed sensor data — the distinction is in how much is applied at record time and how much is left to the facility.

A compressed video codec — H.264, ProRes 422, DNxHR — bakes all ISP decisions in at record time. Demosaicing, white balance, colour space, and tone mapping are applied by the camera before the file is written. The colourist receives footage that already exists in a defined colour space with a baked white balance. RAW defers those decisions: demosaicing, white balance, colour science, and tone mapping are applied in post, using the facility’s tools and the manufacturer’s decode SDK.

The value of deferral is practical. White balance is adjustable with no quality penalty — a 500 K shift in post costs nothing in a RAW workflow, whereas the same correction applied to a baked log file introduces some chroma noise. Exposure latitude is effectively greater, because the facility can adjust the decode ISO before any clipping judgement is made. And the debayering algorithm is chosen by the post facility — for high-end work, this means using the manufacturer’s own premium-quality algorithm rather than whatever the camera applied in real time.

Crucially, the value of RAW is not that it looks better. An identically exposed log clip from a high-quality codec is often comparable in the final grade. The value is in flexibility and reversibility — and in preserving the full sensor capture as a true camera negative for future deliverables.

02 — The Bayer mosaic and demosaicing

Individual photosites on a sensor are monochromatic — each photodiode measures light intensity, not colour. To capture colour, a Bayer colour filter array (CFA) is placed over the sensor surface. Each photosite sees only one wavelength band — red, green, or blue — determined by the filter above it. A RAW file contains this mosaic of single-channel values: one number per photosite, each representing a single colour channel.

The standard Bayer pattern is RGGB — one red, two green, and one blue in each 2×2 block. The reason for doubling green is physiological: human vision is most sensitive to the green portion of the spectrum, which carries the majority of perceived luminance detail. Having two green photosites per 2×2 unit halves the luminance noise and preserves perceived sharpness, at the cost of halving the spatial resolution of each colour channel.

Demosaicing (also called debayering) is the process of reconstructing a full-colour RGB image from this mosaic. For every photosite, the camera or post software must interpolate the two missing colour channels from neighbouring photosites. A red photosite, for instance, has no green or blue value of its own — those are inferred from the surrounding green and blue photosites. The quality of this interpolation has a real and measurable impact: fine detail resolution, colour fringing at high-contrast edges, and moiré in fine repeating textures all depend on the algorithm used.

Demosaicing algorithms

Simple bilinear interpolation — averaging the nearest same-colour neighbours — is computationally fast but produces poor results: colour fringing on diagonal edges, softened fine detail, and visible moiré. It is used for proxy generation and real-time preview only.

More sophisticated algorithms such as AHD (Adaptive Homogeneity-Directed) and similar approaches analyse local image structure to detect edges and adapt the interpolation direction accordingly. These are used in many general-purpose RAW processors and produce significantly better results than bilinear interpolation on most image content.

The highest quality is achieved with proprietary algorithms from ARRI, Blackmagic, and RED, each optimised for their specific sensor architecture and pixel geometry. These are the algorithms deployed via each manufacturer’s SDK in DaVinci Resolve. Resolve exposes debayer quality settings — Premium, Optimised, and lower — for supported RAW formats. Always confirm Premium debayer quality for the final grade; Optimised or lower settings are appropriate for offline and proxy work only.

Beyond Bayer — the Blackmagic RGBW sensor

The URSA Cine 12K LF and URSA Cine 17K 65 use a sensor architecture that adds panchromatic (W — white/clear) photosites to the standard RGB array. W photosites have no colour filter — they capture all visible wavelengths, gathering significantly more light than adjacent filtered photosites. Blackmagic’s processing pipeline uses the W channel to lift the signal-to-noise ratio of the luminance channel, using the additional light data to reduce noise without affecting colour fidelity.

The result is cleaner images at any ISO, with 16 stops of dynamic range on both the 12K and 17K models. This architecture is fundamentally distinct from a standard Bayer pattern — the in-camera processing required to combine W and RGB data before writing the BRAW file is considerably more complex than standard Bayer demosaicing. The BRAW file format itself is identical to that used by other Blackmagic cameras; the RGBW sensor difference is in the hardware and Blackmagic’s decode pipeline, not the file container.

03 — RAW vs log: the key distinction

This is the most common point of confusion in camera-to-post discussions. RAW and log are often presented as competing quality levels — they are not. They are different approaches to the same problem: preserving the sensor’s captured information for post-production. The distinction is about where decisions are made, not how good the result is.

Log is a transfer function: the sensor data has been demosaiced and white-balanced by the camera’s ISP; a log curve is applied to compress dynamic range for recording. The footage exists in a specific, defined colour space — S-Log3/S-Gamut3, Log C4/AWG4, BRAW with a log display transform — and all of those decisions are made at record time, before any post involvement.

RAW defers the decode: the camera records the sensor’s mosaic data with metadata describing how it was captured. Demosaicing, white balance, and colour science decisions are made in post by the facility, using the manufacturer’s decode SDK. Both are legitimate professional approaches — they are not competing quality levels.

04 — RAW formats

Every major camera manufacturer has a RAW format. They differ significantly in compression approach, how much processing is applied before recording, and what the post facility controls. Understanding the practical differences — not just the marketing claims — is essential when specifying a camera package for a colour-critical production.

05 — RAW in a colour-managed pipeline

RAW adds a decode stage upstream of the normal colour management pipeline. In DaVinci Resolve, this is the Camera Raw panel — accessible at the top of the Colour page viewer, above the node graph. Settings here are applied before any node in the graph; they affect the entire grade for every clip to which they are applied.

The Camera Raw panel in DaVinci Resolve

ISO, colour temperature, tint, and exposure are set in the Camera Raw panel. These parameters are applied upstream of the node graph — changing them after a grade is established affects every node downstream. Treat Camera Raw settings as production decisions, not creative tools.

For BRAW, the decode mode can be set to Camera Metadata — which uses what was set on camera at the time of recording — or Custom, which allows full post adjustment of ISO, colour temperature, tint, and exposure. Resolve also exposes Blackmagic’s colour science choices in this panel, including the generation of Blackmagic Design colour science to apply.

For ARRIRAW, ARRI’s colour science is applied by their SDK. The Camera Raw panel exposes the exposure index (EI) and allows selection of the decode colour space — Log C4 / AWG4 for ALEXA 35 material, Log C3 / AWG3 for earlier bodies. For R3D, RED’s IPP2 colour science is applied; ISO and white balance are fully adjustable in the panel.

Scene-linear output for VFX

RAW decodes to scene-linear data in the camera’s native gamut — ARRI Wide Gamut 4, Blackmagic Wide Gamut, REDWideGamutRGB. VFX vendors working in ACES want scene-linear ACES2065-1: the IDT converts from the camera’s decoded linear to the ACES reference space. When sending RAW material to VFX vendors, two approaches are common.

Sending the raw files directly gives the VFX artist maximum flexibility — they decode using the same manufacturer SDK with the agreed settings, and work in their preferred scene-linear space. This requires that the VFX facility has the correct SDK installed and that decode settings are clearly documented and agreed. Sending rendered scene-linear EXR files — in an agreed colour space such as ACES2065-1 or the camera’s native linear — is a simpler handoff that locks the colour science and avoids any decode ambiguity. For controlled VFX pipelines with multiple vendors, rendered EXR is generally preferred.

Archiving RAW

ARRIRAW in ACES2065-1 linear is the Academy’s recommended archival format for ARRI camera originals. BRAW files are self-contained with embedded metadata and are a valid archival format for Blackmagic camera originals — the file contains everything needed to reproduce the grade exactly. Never archive only a rendered deliverable. The RAW file — or the high-quality log original where RAW was not recorded — is the camera negative. A deliverable is a copy made for a specific purpose at a specific time; it is not an archive.

06 — When to shoot RAW

RAW is not always the right choice. The decision is creative, technical, and logistical. For any given production, evaluate the camera’s RAW implementation against its log output before committing to a RAW pipeline — some cameras are more thoroughly optimised for their log output than their RAW decode, and a careful on-camera test is worth doing before a long shoot.