Some BLP members use DxO Photolab for post processing. This may be of interest.

DxO recently released version 4 of DxO Photolab which has added a new version of its excellent noise reduction software (PRIME) called 'DeepPRIME". PRIME noise reduction has been unchanged over previous versions of DxO (to my knowledge at least) and indeed I think was developed before DxO was called 'PhotoLab'. The introduction of DeepPRIME means that DxO Photolab 4 now offers three versions of noise reduction: HQ (which is a relatively low quality NR designed for fast processing of relatively low noise files), PRIME, and now DeepPRIME, each of which is selectable from the menu.

Several software producers have introduced 'artificial intelligence' into their software, and DxO has followed this trend. I don't pretend to understand how this works, but DxO claims the relevant algorithms have been 'trained' by using millions of images to provide an improved noise reduction process. As I understand it, it uses pre-categorised images to work out how to apply the optimal noise reduction to a particular image. Noise reduction is applied during the raw file demosaicing process.

What does this mean in practice? I've been using DxO Photolab since its inception for processing bird images, and have found it's PRIME noise reduction to be very impressive. Generally it finds a very good balance between preserving detail and reducing noise, and is able to recognise and distinguish between (for example) feather detail and background and adjust the level of NR appropriately. For those unfamiliar with it, it is a 'one step' NR applied across the whole image when the file is exported; you don't select different parts of the image for different levels of NR, as you do in some other software. It's effects can only be seen in full on exported files as it is too computationally intensive to apply in 'real time' as you are adjusting a raw image. I've been using a trial version of v4 for a few days now, to process bird photos and also trialling some ISO6400 and above files to see whether DeepPRIME it makes any difference.

My observations so far are that DeepPRIME shows modest but worthwhile improvements on PRIME when used on my Canon 90D CR3 raw files. At or below ISO 800 the differences are quite small. You can see them at 100% in images, but the differences are not really significant in my view. As you would expect, different files show different amounts of difference compared to PRIME with some showing noticeable changes and others showing virtually no visible difference As ISO increases the improvements become a bit more obvious. To my eye these differences are mainly about crispness and cleanness of detail. DeepPRIME seems to detect and resolve edges more cleanly and with fewer artifacts. This leads to (for example) thin details like hairs or feather structure looking slightly 'cleaner' and smoother. DeepPRIME is also more decisive in resolving detail in areas like white or pale patches; it's not finding more detail, but the detail looks slightly less muddy as PRIME still leaves a bit of noise in places. Sharp differences between (say) an object and smooth background (eg the edge of a beak) also seem to be a little smoother. Interestingly, occasionally this makes the resulting section of the image seem a little less detailed and sharp, because having a little noise left in an area can trick the eye into seeing more detail than is actually there (= false high frequency detail).

I have attached a couple heavy crops of ISO 12800 files taken indoors in available light using the 90D. Pixel peepers will notice the effect that I mention above - the details of the hairs are slightly crisper and smoother with DeepPRIME, the 'iris' of the toy's eye is smoother and shows fewer artifacts. However, this is at a huge ISO for a high megapixel APSC camera. Viewed at (for example) full screen on an HD monitor, or downsized for BLP, or printed at modest sizes, these differences will be less obvious. Of course, no noise reduction software will overcome things like reduced dynamic range from high ISO files, even if reducing visible noise.

The downsides are computational; files exported with PRIME are already slow on my modest PC and DeepPRIME makes the process even slower, particularly where the files are full of detail (eg a leafy background). My CPUs fan can be heard working overtime to cool the CPU and Windows shows the processer continuously operating at 100%. However, this slowness is only on exporting files; adjusting things like lighting, colour etc is fast and unchanged with the new version. DxO is definitely software which is processor-limited and adding RAM above 16GB does not improve speed in my experience.

The new software includes some other worthwhile additions. Users of DxO will immediately notice that the different development features are now better organised; version 3 has all functions in a continuously scrolling list on the right, whereas v4 now organises functions into separate tabs (eg lighting, detail, colour) each dealing with a limited set of functions. I understand these are also customisable. v4 also includes watermarking which will be very useful to those who use it, making it a more complete 'one size fits all' product. The new version continues with powerful local adjustment functions (based on Nik technology) and DxO's surprisingly useful colour wheel for adjusting separate colour tones.

Note that both PRIME and DeepPRIME are only applied to raw files, because (as I understand it) the noise reduction is done at the same time as demosaicing. Those who like to produce TIFF files from raw (eg using Canon DPP), and then process the TIFFs for noise and sharpness, would only be able to use the HQ noise reduction. DxO also does not work with DNG files.

V4 must be purchased, either as an upgrade, or as a new product for those who don't already own DxO Photolab.

Simon