Understanding EXR Data Compression

Information compiled by Nicolas Dufresne for Rainbox Productions and Rainbox Laboratory. Contact us if you see any mistake or have suggestions for
this document. https://rainboxlab.org
Many thanks to Dino Muhić for his expertise about DWA.

When using the OpenEXR format, it may be hard to determine what is the best data compression to use, and the question is raised very often. This
document is an attempt to answer it, depending on what is your need with the file and what type of image you’re storing.

A compression is said to be the best for a specific need when its read or write speeds better fit the application, and the file size is the smallest available.

Types of images
In this document, we’re distinguishing between the following different types of images:

Grainy video or animation

Photographic images (including realistic CGI) or animation with grain

Video
Photographic images (including realistic CGI) without grain

Animation, Graphics
Highly stylized images such as 2D Animation, Motion Graphics, stylized 3D Animation, or 3D passes such as Z-Depth, Normal...

Solid colors, large flat areas

Images mostly consisting of solid colors, such as alpha or id channels

Texture maps
Multi-Resolution files

Summary
Lossy (final) exports
Especially for final exports, you probably don’t need lossless compression, and with the right settings, the exported file size can be quite small with just
a very small reduction of quality, and the Master or backup of your movie can be stored this way.

In this case, DWAA (with a small compression level) is very efficient in all cases. Be careful, DWA is not supported by ffmpeg (yet). In this case
PXR24 is a good (lossless) alternative.
Solid colors (e.g. alpha channels) can be compressed very efficiently using RLE without losing quality.

If you’re not exporting for complex compositing (e.g. choma keying), and especially from a video source or if the export is to be used as a Master for
further YUV 422 or 421 video exports (like h.264), the Luminance/Chroma option will divide the file size by two with only a very small reduction of
quality. Be careful, ffmpeg does not support the Luminance/chroma option (yet).

Lossless (intermediary) exports

If the file is to be used in a compositing software for example, you may want to export without losing quality.

In any case, if your rendering an image with AOV (most likely from a 3D software), and you can accept a (very small) quality loss, DWA is the best
option, as it will compress only the RGB channels (or Y, RY, BY in case of Luminance/Chroma) and use RLE for alpha and ZIP for any other channel.
In this case, be careful that DWA uses the channel names (case sensitive):
• Lossy channels: R, G, B, Y, RY, BY
• RLE: A
• ZIP: any other name (Red, red, r, Green, green, g, Blue, blue, b, x, y, z, U, u, V, v, etc…)
Be careful with the names of the channel. For example using XYZ will result in the Y channel being lossy. You can use "xyz" instead.

When the images don’t have grain:

If you don’t need full 32-bit float precision, PXR24 is the best compression you can use.
If you need full 32-bit float precision, ZIP is the best option.

When the images have grain, PIZ is always the best option.

For stereo images, the best is ZIP.

For solid colors such as alpha channels, it’s better to use RLE.

Special cases
There are more specific uses for OpenEXR:

On systems for real-time playback, B44 is preferred (or B44A for alpha channels and solid colors).

If you need to render small-size lossy proxies, you can use DWAA with a high compression level.
Gray-scale images will benefit a lot from the Luminance/Chroma option, without losing quality.

Luminance/Chroma images
Encoding flat images with one luminance and two chroma channels, rather than as RGB data, allows a simple but effective form of lossy data
compression that is independent of the compression methods listed here. The chroma channels can be stored at lower resolution than the luminance
channel. This leads to significantly smallerfiles, with only a small reduction in image quality.

This is the same kind of method as what is called YUV 422 for video.

That means you can use this option if you’re exporting to or from a standard YUV 422 or 421 video without losing quality.
If you’re exporting to or from YUV 444 or RGB, the reduction in quality is still very small.

As luminance is stored in full quality using the Luminance/Chroma option, it can store gray-scale image much better than standard RGB without losing
any data.

Note : this option is not supported by ffmpeg (yet).

Formats you don’t know you already know

ZIP is the compression method used in Portable Network Graphics (.png) files. It is NOT based on the Zip file format, despite its name). That means
when this document states you can use ZIP, it can be replaced by PNG files (if you don’t need AOV, keeping in mind that most software is slower
handling PNG, especially After Effects).

RLE is a compression used in TGA files, and it is close to the way data is compressed with Quicktime Animation sometimes used in .mov files. This
means you can replace RLE by TGA files (if you don’t need AOV)

DWAA is close to the way data is compressed in JPEG files. DWA could be replaced by a JPEG file sequence, if you’re careful about the compression
rate (and you don’t need alpha or AOV).

Summary by usage and type of image

• Final render/export

Animation, Graphics or Video

DWA, (a bit) lossy (use a small compression level) – not supported by ffmpeg (yet).
PXR24, lossless*
If exporting to YUV 422 or 421 video (e.g. h.264) or gray-scale, use the Luminance/Chroma option – not supported by ffmpeg (yet).

Grainy video or animation

DWA, a bit lossy – not supported by ffmpeg (yet).
PIZ, lossless
If exporting to YUV 422 or 421 video (e.g. h.264) or gray-scale, use the Luminance/Chroma option – not supported by ffmpeg (yet).

Solid colors, large flat areas (alpha and id channels)

RLE

* In the specific case of a final export, PXR24 is considered lossless, as there should not be any need for 32-bit float data.

• Intermediary 32-bit float

Texture maps, Animation, Graphics or Video

ZIP
DWA if you can afford a small quality loss – not supported by ffmpeg (yet).
Note that DWA will only compress R,G,B channels (or Y,RY,BY in case of Luminance/Chorma) and automatically select either RLE
for alpha or ZIP for AOV (Z, U, V, Normal...).

Grainy video or animation

PIZ

Solid colors, large flat areas (alpha and id channels)

RLE

• Intermediary 16/24-bit float, 16/32-bit int

Texture maps, Animation, Graphics or Video

PXR24, if unavailable: ZIP
 DWA if you can afford a small quality loss – not supported by ffmpeg (yet).
Note that DWA will only compress R,G,B channels (or Y,RY,BY in case of Luminance/Chorma) and automatically select either RLE
for alpha or ZIP for AOV (Z, U, V, Normal...).

Grainy video or animation

PIZ

Solid colors, large flat areas (alpha and id channels)

RLE

• Stereo images

Texture maps, Animation, Graphics or Video

ZIP

Solid colors, large flat areas (alpha and id channels)

RLE

• Real-time playback

B44A or B44
If unavailable: PXR24

• Proxies

DWAA (use a high compression level)

If unavailable: PXR24, ZIP, PIZ
Name Loss Description* Ratio** Summary Good for***
PIZ Lossless A wavelet transform is applied to the pixel 35~55% - Write speed = read speed - Photo/Video (with grain)
data, and the result is Huffman-encoded. (photo with - Best for grainy images
This scheme tends to provide the best grain) - 3D Animation (with
compression ratio for the types of images grain)
that are typically processed at Industrial
Light & Magic. Files are compressed and
decompressed at roughly the same speed.
For photographic images with film grain,
the files are reduced to between 35 and 55
percent of their uncompressed size. PIZ
compression works well for scan-line based
files, and also for tiled files with large tiles,
but small tiles do not shrink much. (PIZ-
compressed data start with a relatively
long header; if the input to the compressor
is short, adding the header tends to offset
any size reduction of the input.)

PIZ compression is only supported for flat

images.

ZIP Lossless Differences between horizontally adjacent pixels 45~55% - Faster reading, significantly Only when 32bpc float is
are compressed using the open-source zlib (photo slower writing needed (otherwise,
library. ZIP decompression is faster than PIZ without grain) - Same as PNG PXR24 is better):
decompression, but ZIP compression is - Supported for stereo images
significantly slower. Photographic images tend to - Texture maps
shrink to between 45 and 55 percent of their
uncompressed size. Multi-resolution files are - Photo/Video (without
often used as texture maps for 3D renderers. For grain)
this application, fast read accesses are usually
more important than fast writes, or maximum - 3D Animation (without
compression. For texture maps, ZIP is probably grain)
the best compression method. In scan-line based
files,16 rows of pixels are accumulated and - 2D Animation, Graphics
compressed together as a single block.

ZIPS Lossless Uses the open-source zlib library for - Same as PNG
compression. Like ZIP compression, but - Supported for stereo images
operates on one scan line at a time.

RLE Lossless Differences between horizontally adjacent pixels 60~75% - Fast - Solid colors, large flat
are run-length encoded. This method is fast, and (photo) - Same as TGA areas (alpha and id
works well for images with large flat areas, but - Better with large flat areas channels)
for photographic images, the compressed file (alpha and id channels)
size is usually between 60 and 75 percent of the - Supported for stereo images
uncompressed size.

PXR24 Lossless After reducing 32-bit floating-point data to 24 bits Better than - Faster reading, significantly Only for 16bpc float or
(16bit float, by rounding(while leaving 16-bit floating-point ZIP slower writing 16/32bpc int, or when
16/32-bit data unchanged), differences between for 32bpc - Turns 32-bit float to 24-bit 24bpc float is sufficient
int) horizontally adjacent pixels are compressed with instead of 32bpc
zlib, similar to ZIP. PXR24 compression Same as ZIP
preserves image channels of type HALF and otherwise - Photo/Video (without
Slightly UINT exactly, but the relative error of FLOAT grain)
Lossy data increases to about . This compression
3x10-5 method works well for depth buffers and similar - 3D Animation (without
(32-bit images, where the possible range of values is grain)
float) very large, but where full 32-bit floating-point
accuracy is not necessary. Rounding improves - 2D Animation, Graphics
compression significantly by eliminating
thepixels' 8 least significant bits, which tend to be - Texture maps
very noisy, and therefore difficult to compress.
PXR24 compression is only supported for flat
images.

B44 Lossy Channels of type HALF are split into blocks of 44% - Fixed file size - For systems needing
four by four pixels or 32 bytes. Each block is then - Very fast read speed real-time playback
packed into 14 bytes, reducing the data to 44
percent of their uncompressed size. When B44
compression is applied to RGB images in
combination with luminance/chroma encoding
(see below), the size of the compressed pixels is
about 22 percent of the size of the original RGB
data. Channels of type UINT or FLOAT are not
compressed. Decoding is fast enough to allow
real-time playback of B44-compressed OpenEXR
 image sequences on commodity hardware. The
size of a B44-compressed file depends on the
number of pixels in the image, but not on the
data in the pixels. All images with the same
resolution and the same set of channels have the
same size. This can be advantageous for
systems that support real-time playback of image
sequences; the predictable file size makes it
easier to allocate space on storage media
efficiently. B44 compression is only supported for
flat images.

B44A Lossy Like B44, except for blocks of four by four pixels < 44% - Very fast read speed - For systems needing
where all pixels have the same value, which are - Better with large flat areas real-time playback
packed into 3 instead of 14 bytes. For images (alpha and id channels)
with large uniform areas, B44A produces smaller
files than B44 compression. B44A compression
is only supported for flat images.

DWAA Lossy JPEG-like lossy compression format contributed Varying - Same as JPEG - Proxies (high
by DreamWorks Animation. Compresses 32 depending on compression)
scanlines together. chosen
compression - Final exports when a
level small compression is
acceptable

DWAB Lossy Same as DWAA, but compresses blocks of 256 Varying - For proxies
scanlines. depending on
chosen - Final exports when a
compression small compression is
level acceptable

* The description is taken from the official technical introduction from https://www.openexr.com
** Compressed / uncompressed. A lower ratio is better. For example, 45% means the compressed file size is 45% of the
uncompressed size.
*** A file is good for a given use when its read or write speeds best fits how it is going to be used, and has the best
compression ratio available.