Print Page | Close Window

Looking for high ISO A77 RAW

Printed From: Dyxum.com
Category: Equipment forums
Forum Name: Digital Darkroom
Forum Description: For post-processing software and hardware discussions
URL: https://www.dyxum.com/dforum/forum_posts.asp?TID=95123
Printed Date: 09 October 2024 at 20:47


Topic: Looking for high ISO A77 RAW
Posted By: FineArt
Subject: Looking for high ISO A77 RAW
Date Posted: 18 January 2013 at 08:32
I am looking for a link to a high ISO RAW file for testing. I am waiting for FF nex or A77ver2. I think I have a process for wiping out high ISO noise. If I can make the A77 with tiny pixels behind a pellicle work I might go that route.

Of course the file cannot be something important to you because you would have to provide a link on the net for download. Once its on the net it is probably being used somewhere. An average shot that is lost from high ISO will be fine. 3200 or 6400 with good focus, natural complex textures would help testing (why make it easy?) ETTR. Clipped highlights are fine. Please no badly blown or badly underexposed file. A DNG would be good in case my converter cant open newer arws.

If on the other hand you have a shot you want recovered I can send you back the jgp link via PM so the world doesn't use it. PM me when you have it so I can delete from sendspace.com

I have no rights to your picture. I just want something for testing. It will not be used again for anything. I will probably post a screenshot crop from a few raw converters. Nothing near a full size image.

Thanks.



Replies:
Posted By: MediaArchivist
Date Posted: 18 January 2013 at 11:10
Here's a few high ISOs shot in RAW mode on my a77. I converted to DNG, they range from 4000 to 8000:
http://mediaarchivist.s3.amazonaws.com/share/sample_dng/001_ISO%206400.dng" rel="nofollow - 6400 ISO
http://mediaarchivist.s3.amazonaws.com/share/sample_dng/002_ISO%206400.dng" rel="nofollow - 6400 ISO
http://mediaarchivist.s3.amazonaws.com/share/sample_dng/003_ISO%204000.dng" rel="nofollow - 4000 ISO
http://mediaarchivist.s3.amazonaws.com/share/sample_dng/004_ISO%204000.dng" rel="nofollow - 4000 ISO
http://mediaarchivist.s3.amazonaws.com/share/sample_dng/005_ISO%208000.dng" rel="nofollow - 8000 ISO

Enjoy!

-------------
α99ii/α7Rv • lenses used incorrectly ➜ http://www.roxplosion.com/ - roXplosion!


Posted By: FineArt
Date Posted: 18 January 2013 at 16:13
Thanks very much, downloading the first one now.


Posted By: FineArt
Date Posted: 18 January 2013 at 16:24
What version of DNG is that? On the first one when I do a conversion it is scrambled. I use adobe dng converter on it again to covert to "compatible camera raw 5.4 and later" it gives me a black and white scene of a band stage. There is no noise in it.


Posted By: FineArt
Date Posted: 18 January 2013 at 16:29
The ISO 8000 file looks normal. A fisheye in a bar. I will try that one.


Posted By: FineArt
Date Posted: 18 January 2013 at 21:08
After brightening the file it had strips that did not line up. Maybe you have a newer dng converter that does not use baseline tiff.

I got an ISO 6400 from member retyred today. No lower ISO versions. I have taken MS Paint screenshots with faces in a public place cut out.
These are linked due to 1920x1080 screensize.

Here is the shot in IDC
http://farm9.staticflickr.com/8235/8392362787_b4d569a695_o.jpg" rel="nofollow - http://farm9.staticflickr.com/8235/8392362787_b4d569a695_o.jpg

Here is the shot as converted (linear, the same as the sensor reads it) in Images Plus
http://farm9.staticflickr.com/8374/8392362827_e80fed8065_o.jpg" rel="nofollow - http://farm9.staticflickr.com/8374/8392362827_e80fed8065_o.jpg

Here is Images Plus with curves, NR, Sharpening
http://farm9.staticflickr.com/8221/8392362847_049c8f4967_o.jpg" rel="nofollow - http://farm9.staticflickr.com/8221/8392362847_049c8f4967_o.jpg

Here is IDC 100% crop
http://farm9.staticflickr.com/8336/8392380995_fe5bac5187_o.jpg" rel="nofollow - http://farm9.staticflickr.com/8336/8392380995_fe5bac5187_o.jpg

Here is Images Plus 100% crop
http://farm9.staticflickr.com/8229/8392380957_f0d8b88296_o.jpg" rel="nofollow - http://farm9.staticflickr.com/8229/8392380957_f0d8b88296_o.jpg

This could have been improved further with more time to adjust routines.

To my eye this makes A77 ISO6400 workable if you are in low light. Some watercolors are to be expected at this ISO. There is still enough detail in a 24mp shot to make it work.


Posted By: MediaArchivist
Date Posted: 18 January 2013 at 22:24
I'm sorry my files did not work out. Just for curiosity's sake, I used ARW files from my a77, and used LightRoom 4 to export them as DNG. The export settings were "Camera Raw 5.4" and "Embed Fast Load Data". I should also mention that these are the first DNGs I ever made, so I don't know if there is some tweak I was missing or accidentally had set.

The results you published look very promising. Is this software you plan to release at some point?

-------------
α99ii/α7Rv • lenses used incorrectly ➜ http://www.roxplosion.com/ - roXplosion!


Posted By: FineArt
Date Posted: 18 January 2013 at 23:07
Dont worry about it. DNG converter has gone through many versions. If I had used a newer version the other software might not have recognized it.

The other software is available from a mathematician. Images Plus. He designed it for astronomy. He has many camera control functions for Canon and Nikon. Unfortunate he has not done as much to support Sony. It will open raws from sony depending on how new the camera is.

The big advantages are it opens the file as linear data exactly as the camera records it +de-bayering. You can apply whatever curve you want rather than the standard gamma 2.2 that other software does automatically. It has extensive noise and sharpening routines. It takes a while to learn to use all the features.

It will convert to 32 bit FP or 64 bit integer. Multiple files for HDR can be put in such a big space.

Noise reduction is mostly about smoothing. Doing math functions in FP really wipes out the accumulation of rounding errors as you manipulate the file. The same for sharpening.

The big trick I discovered using this software is treating your file as 12 dimensions. You can split the image into 4 32bit tifs with 3 dimensions. RGB and L (luminance) each with 2 dimensions of space and 1 of value. Run each through a good noise program as a B/W tiff. Optimize the NR for that file. Then recombine to RGB with Images plus.

From that point you will have a tiny amount of noise. Understand that you have coordinated the NR to smooth each channel. Other programs have not figured this out. Their results would be much better if they had. Use standard NR and sharpening on the combined file.

Do everything in 32 bit FP if you can.

This process will let Sony SLTs at high ISO run with FF Canon or Nikon.



Posted By: slawrencephoto
Date Posted: 19 January 2013 at 00:10
Originally posted by FineArt FineArt wrote:

I am looking for a link to a high ISO RAW file for testing. I am waiting for FF nex or A77ver2. I think I have a process for wiping out high ISO noise. If I can make the A77 with tiny pixels behind a pellicle work I might go that route.

Of course the file cannot be something important to you because you would have to provide a link on the net for download. Once its on the net it is probably being used somewhere. An average shot that is lost from high ISO will be fine. 3200 or 6400 with good focus, natural complex textures would help testing (why make it easy?) ETTR. Clipped highlights are fine. Please no badly blown or badly underexposed file. A DNG would be good in case my converter cant open newer arws.

If on the other hand you have a shot you want recovered I can send you back the jgp link via PM so the world doesn't use it. PM me when you have it so I can delete from sendspace.com

I have no rights to your picture. I just want something for testing. It will not be used again for anything. I will probably post a screenshot crop from a few raw converters. Nothing near a full size image.

Thanks.


What cameras do you currently use for your photography?

Si

-------------
Life, The Universe & Everything
http://slawrencephotography.com/ - My Website
RX100, Sigma 12-24mm F4.5/5.6 DG HSM, Sony 24-70mm ZA, Sigma 70-200mm F2.8 APO DG EX OS


Posted By: FineArt
Date Posted: 19 January 2013 at 00:48
A55, A350


Posted By: rickztahone
Date Posted: 19 January 2013 at 02:07
Originally posted by FineArt FineArt wrote:

Dont worry about it. DNG converter has gone through many versions. If I had used a newer version the other software might not have recognized it.

The other software is available from a mathematician. Images Plus. He designed it for astronomy. He has many camera control functions for Canon and Nikon. Unfortunate he has not done as much to support Sony. It will open raws from sony depending on how new the camera is.

The big advantages are it opens the file as linear data exactly as the camera records it +de-bayering. You can apply whatever curve you want rather than the standard gamma 2.2 that other software does automatically. It has extensive noise and sharpening routines. It takes a while to learn to use all the features.

It will convert to 32 bit FP or 64 bit integer. Multiple files for HDR can be put in such a big space.

Noise reduction is mostly about smoothing. Doing math functions in FP really wipes out the accumulation of rounding errors as you manipulate the file. The same for sharpening.

The big trick I discovered using this software is treating your file as 12 dimensions. You can split the image into 4 32bit tifs with 3 dimensions. RGB and L (luminance) each with 2 dimensions of space and 1 of value. Run each through a good noise program as a B/W tiff. Optimize the NR for that file. Then recombine to RGB with Images plus.

From that point you will have a tiny amount of noise. Understand that you have coordinated the NR to smooth each channel. Other programs have not figured this out. Their results would be much better if they had. Use standard NR and sharpening on the combined file.

Do everything in 32 bit FP if you can.

This process will let Sony SLTs at high ISO run with FF Canon or Nikon.



I must admit that most of what you said is over my head but I hope you are right about what you are talking about. Working these files in this fashion seems like it would really help improve our NR. Subscribed.

-------------
a99+VG|a77+VG|a55|Nex6|HVL-56/58|minO|58 1.2|24|Tam|90|SAL||16-50|70-200|∑|50 1.4|∑| 24-70 2.8
[URL=http://www.flickr.com/photos/rickztahone/]Flickr


Posted By: FineArt
Date Posted: 19 January 2013 at 02:30
Here is the red channel. The noise that you see as bright red and green spots is from brights and darks on the individual channels.

Noise reduction on these has profound effect on the combined file.

http://farm9.staticflickr.com/8050/8393076997_dcc4246812_o.jpg" rel="nofollow - http://farm9.staticflickr.com/8050/8393076997_dcc4246812_o.jpg


Posted By: FineArt
Date Posted: 19 January 2013 at 03:01
How's about some color blotches like Raw converters show with NR off? I made them myself! Not by turning NR off but by using strong smoothing on the individual channels.

http://farm9.staticflickr.com/8356/8393128211_bdb81f6ab8_o.jpg" rel="nofollow - http://farm9.staticflickr.com/8356/8393128211_bdb81f6ab8_o.jpg

I can make them square, rounder, all kinds of patterns by adjusting smoothing routines. You cannot create them with random noise on a sensor. Noise from your sensor will be random specks, bands, or an amp glow spot (heat). Any time you see an artificial looking pattern it is made by your software.


Posted By: FineArt
Date Posted: 19 January 2013 at 16:29
I did some testing of the Nikon D800 with this method. ISO 3200 looks like ISO200 It's basically noise free. ISO 12800 (2x 6400 which I have never used in the real world) retains almost as much detail as a base ISO shot. There is a fine noise pattern in chroma and luminance that is not objectionable. As long as it looks like fine pixels of random noise it looks good.

That brings me to the reason for using this software. It is for scientific use when people need to get the most out of very faint objects. Some deep sky objects may need hrs of exposure to bring them out. Noise mitigation has to be top notch. Actually all you need is the purest conversion of what the sensor provides as possible.

I did the tests on the D800 because raws were available on Imaging resource. They have not provided A77 raws. Sony A99 conversion should be very close to the quality of noise free at ISO6400.

As the ISO climbs you can see the reduction in dynamic range.

Most people will want to use their favorite RAW conversion software instead. Be warned that if the program accumulates noise into clumps, NR on those by channel, will lead to larger lumps of color noise. The reason is the smoothing will not be aligned by color. You will see the color patterns you see in many raw conveters. Its crucial that the noise stay as fine and as random as possible.

Raw Therapee with everything off looks very similar to Images plus. It is a clean conversion. Amaze will mike fine lines render a hair narrower at the expense of a faint artificial look to the image relative to Images plus.

Sony IDC also leaves the noise as fine random pixel level jumps in value.

Both of these will output 16 bit tif which you can use in a program that separates channels. Export to NR then back to 16 bit tif. The advantage is to Images plus that the whole thing can be done in 32 bit fp. It has extensive NR routines including several that preserve fine detail.

You can split channels 3 ways RGBL, CMYL, HSL.
If you go from the original to these splits, Do NR, recombine, then save, you will have an extremely low noise file. If you then average all three methods you have basically a zero noise file. Your file becomes limited by the DR of the sensor. Yes, it's a fair amount of work to do this. Some files are important enough to go through the trouble.

One last thing, avoid any NR smoothing in luminance channel. You must use detail preserving techniques. On the color channels it is less critical. Use pixel smoothing if you have to. Know that it will result in some color lumps in your final file.


Posted By: FineArt
Date Posted: 19 January 2013 at 16:37
This was the JPG output of ISO3200
http://fs05n5.sendspace.com/dl/b6ee75e0ce6682a4ecb9a9d059e5e1ce/50facb6d1781b996/w1tq12/CVT_D800hSLI03200NR0%20curves.jpg" rel="nofollow - http://fs05n5.sendspace.com/dl/b6ee75e0ce6682a4ecb9a9d059e5e1ce/50facb6d1781b996/w1tq12/CVT_D800hSLI03200NR0%20curves.jpg


Posted By: FineArt
Date Posted: 19 January 2013 at 19:27
I just downloaded Sony A99 RAW ISO6400 from imaging resource. I did a RGBL split, NR, sharpen, re-combine.

Here are 3 100% crops. Compare them to the same file jpg at Imaging resource. Its a big difference.

http://farm9.staticflickr.com/8357/8395063175_197342b41d_o.jpg" rel="nofollow - http://farm9.staticflickr.com/8357/8395063175_197342b41d_o.jpg


http://farm9.staticflickr.com/8495/8395063227_6dba0be2d9_o.jpg" rel="nofollow - http://farm9.staticflickr.com/8495/8395063227_6dba0be2d9_o.jpg


http://farm9.staticflickr.com/8090/8395063303_06438f6c3e_o.jpg" rel="nofollow - http://farm9.staticflickr.com/8090/8395063303_06438f6c3e_o.jpg

50% crop
http://farm9.staticflickr.com/8043/8395095985_82f570f8f8_o.jpg" rel="nofollow - http://farm9.staticflickr.com/8043/8395095985_82f570f8f8_o.jpg


ISO6400! I'll call this Arthur's method.

Attached is a screenshot of the red channel before and after NR. Each channel has it's own noise.

http://farm9.staticflickr.com/8332/8397000750_8eda888755_o.jpg" rel="nofollow - http://farm9.staticflickr.com/8332/8397000750_8eda888755_o.jpg


Posted By: FineArt
Date Posted: 22 January 2013 at 01:49
I just got a reply over at LuLa that Bruce Fraser published by channel noise reduction and sharpening many years back.

http://www.luminous-landscape.com/forum/index.php?topic=74352.0" rel="nofollow - http://www.luminous-landscape.com/forum/index.php?topic=74352.0

So it's been done and it works.



Print Page | Close Window