READ: A77 true max ISO and something disturbing about A77 RAW

Started Oct 22, 2011 | Discussions thread
ForumParentFirstPreviousNextNext unread
Flat view
TimberWolfQX
Regular MemberPosts: 218
Like?
READ: A77 true max ISO and something disturbing about A77 RAW
Oct 22, 2011

Hello everyone,

Previously, the A700 was tested to have a max ISO of 1600 (beyond that, the camera simply performed digital multiplication after the analog-digital conversion.

In similar fashion, I decided to test the A77's true RAW ISO because going past that limit will not gain you anything, and may cause highlights to clip instead.

I shot a test scene with A77 FW1.03, RAW mode at various ISOs. The ARW files were converted to DNG and processed using Photobola Raw Analysis Version 2.10.4.0. Sadly, its creator, GaborSch has since passed away.

Picture 1: Linear RAW histogram before interpolation for ISO 100, 1000, 1250, 1600, 3200 and 12800

Findings: The A77 has a 12-bit ADC, and the output is stretched to 14-bit space. This expansion (4096 => 16536 levels) results in 1 level every 4 levels, as can be seen in the ISO 100 histogram. There is a curious dip every 16 levels; I do not know why.

As can be seen from the graphs, the true max ISO of the A77 is 1000 . Past that, you will see that combing of the histogram becomes more and more apparent, and that by ISO 12800, there are very few levels left.

This is in line with the 2/3 stop light loss from SLT Mirror; a previous ISO 1600 amplification is now ISO 1000 after accounting for the light loss.

Recommendation : Shoot with ISO 1000 and underexpose for higher ISOs.
ISO 1600: -2/3 stop
ISO 3200: -1 2/3 stop

Picture 2: Disturbing fact about Sony A77 RAW files: They are NOT RAW

The A77 RAW file exhibits a decompression curve as shown previously by Iliah Borg. I examined the ISO 100 file and determined that Iliah was correct, and that RAW compression was lossy, and in a bad way.

Interpretation of graph: The X-axis represents the data level encoded in the RAW file, and the Y-axis represents the output level after decompression. The gradient of each segment represents the ratio of input to output levels. At the start, from x= 0 to 2048, the gradient is shallow (which is good) as it represents a 1:1 input:output mapping. However, as we go to the brighter parts (higher x values), the gradient increases, showing that fewer levels are available in the highlight part of the image. The histogram crops below show the very drastic reduction in levels available at those thresholds.

Implications:

1. RAW file is NOT RAW: The data here shows that Sony have clearly manipulated the RAW file. A truly RAW file would have a constant ratio mapping of input to output levels (i.e. the graph would have a simple straight line instead of a curve). Whether they did this to save space or not, it is clearly wrong. This a lossy, nonlinear mapping of sensor data that cannot be recovered.

2. They did this in a really stupid way: Throwing away highlight levels means that highlight recovery will not work as well as there are simply insufficient levels to pull back the highlights.

3. Expose to the Right (ETTR) will not work: You are not gaining any finer gradation or tonality by doing so, as the additional levels gained by exposing right will not be recorded properly. Also, I am doubtful if the signal-to-noise ratio will be improved.

Conclusion

The Sony A77 RAW file functionality has serious flaws in implementation, and while I love my A77 for what it does, this is one area that urgently needs addressing.

Thanks for reading!
QX

ForumParentFirstPreviousNextNext unread
Flat view
Post (hide subjects)Posted by
ETTRNew
ForumParentFirstPreviousNextNext unread
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow