Archives (page 2 of 22)

Bangkok – Erewan Shrine

Erewan Shrine, Bangkok


A photograph of the the dancers at the Erewan Shrine is a given for anyone visiting Bangkok and on this day every performance was photographed and videoed by crowds of locals and tourists. I took a few shots as well but my favourite was this one, taken of the women during a rare break in their performance. I liked the contrast of the fabulous costumes and the graceful dancing on the one hand, and on the other the ‘everydayness’ of their activities and posture in this moment. Visitors to the shrine pay for the women to dance and sing prayers to accompany the worshippers’ own prayers and offerings, and on a busy day the women get little chance to take a break.

Abstract

Ivan Vazov Street, Sofia

An old metal gate taking on the appearance of abstract art with many years worth of graffiti and fragments of posters layered and torn.

DXO Deep Prime

Software powered by AI or machine learning appears to be the coming wave in image processing. Skylum – makers of Aurora and Luminar – Topaz Labs and Corel all emphasise the role of AI in their recent advertising and Adobe, with rather less fanfare, is developing and expanding its library of ‘neural filters’ for Photoshop.

My knowledge of how this technology works is close to zero but as far as my limited understanding goes what makes these apps different from ‘traditional’ approaches to processing is that they are ‘trained’ through analysis of vast numbers of images – Adobe claims their Super Resolution feature is trained on millions of images. (As an aside, this suggest to me that the only ‘intelligence’ at work here is the human intelligence of those creating this technology and that ‘machine learning’ is a much more apposite term.)

Many of the AI powered features highlighted in recent advertising appear to function like slightly more sophisticated versions of existing filters, or are designed to process images in ways that I have no interest in – creating flawless skin, replacing or reshaping eyes, or dropping in a completely new sky. Beyond the headline features, though, machine learning is creating some genuinely interesting and useful capabilities. I’ve already assessed one of these in two previous posts on Adobe’s Super Resolution, and a couple of days ago I downloaded a trial version of DXO’s Deep Prime, an AI powered denoising utility which is part of their PureRAW stand alone program. Here are my impressions.

First, I found a nice noisy photograph. The picture was taken in the evening by the B&O canal in Georgetown in Washington DC. While I intended to work from the original untouched RAW file I did eventually decide to boost the exposure because the original was very dark. The picture was taken with a Sony Rx100 at 1/6th of a second and f4 while stabilised on the handrail of the bridge I was standing on to get the shot. ISO was 3200. The image below is a JPEG exported from the Sony ARW file with no adjustments other than the boost to the exposure and Lightroom’s default colour noise and sharpening settings turned off. As you can see there is plenty of noise in the image. (For a closer look in each case right click on the image and ‘open image in new tab’.)

From the original Sony RAW file

Bringing the file (or multiple files) into DXO PureRAW is straightforward though you can’t do it directly from within Lightroom. Either open PureRAW and click on ‘add photos’ or right click directly on any photo in File Explorer and select ‘open with’ DXO PureRAW. The interface is extremely simple:

Once the photo is loaded click to select the photo/s to be processed, click on ‘process photos’ and choose some basic options from the dialogue box. There are two other denoising options available but I can’t think of any reason why you would choose them over Deep Prime. You can also choose to output the processed file as a JPEG rather than DNG, but again I don’t know why you would do that since the whole point of the feature is that you can denoise using DXO technology and still complete the rest of your processing in Lightroom. Choose your detination folder and click on ‘process’. On my laptop (6th gen. i7 2.6Ghz, 16GB RAM, GTX 960M GPU) processing took around 1 minute and 50 seconds and the original 20MB ARW file ended up as a near 72MB DNG file. (I believe the size increase is down to the output file being a linear DNG file, but needless to say I don’t understand what that means.) Once processing is complete you have an ‘export to…’ option which brings up a list of installed compatible editing software.

Make your choice, click on ‘export’ and DXO will launch Lightroom and take you to the import dialogue.

The key question of course is whether it is any good. My answer is a qualified ‘yes’, but I will let you judge for yourself. Below is a JPEG from the DNG created by Deep Prime after processing. I think the programme does an excellent job of removing noise, though it does introduce some softness. That said, I don’t think I could get close to this level of denoising with Lightroom, and certainly not without introducing considerably more softness in the image.

From DNG after processing in DXO Deep Prime

Fortunately, PureRAW has another trick up its sleeve which draws on DXO’s more conventional processing for lens correction. The programme downloads DXO lens correction modules and applies these to the image when processing. If you don’t have the appropriate lens module installed you will be asked if you want to download it when you add a new image. In my sample image, applying both the lens module and the denoising produces a largely noise free image which retains, or in this case exceeds, the sharpness of the original. The difference between the images with and without lens correction is clearly visible, particularly in the structure of the crane and the text on the banners attached to it. That said, further editing in Lightroom does require careful use of the sharpening, texture and clarity sliders after lens correction has been applied since I think the lens correction is something of an over-correction. (Again, right click and select ‘open image in new tab’ for a closer look.)

From DNG after processing in DXO Deep Prime with lens module corrections

So, why a qualified yes? Two reasons – one minor, one major. The minor one is cost. In the current offer period which runs until the end of May you can get DXO PureRAW for $89.99 / €89.99 / £79.99 but after that it reverts to the full price of $129 / €129 / £115. That’s quite expensive. For comparison, Topaz Labs standalone AI tools are mostly between $80-100 full price. Given that PureRAW is clearly aimed at the vast number of Lightroom users I suspect they could drop the price and sell many more licences than they will at the current price.

The major reason is that DXO PureRAW does not work with Fujifilm X-Trans sensors. It is possible to add a Fujifilm RAF X-Trans file and run the programme but the result is an image that looks like it was taken with a lens smeared with grease, as with the image below shot with a Fujifilm X-E2 at ISO 6400. Since processing this file – which is more than 50 percent bigger than the RX100 file – only takes around ten seconds I imagine it is just running it through some bog standard, default noise reduction process.

I have seen some reviews expressing a preference for more control over how the denoising is applied. I can understand why, but I’m happy to have something that doesn’t require yet more choices and decisions from me before I run it. If anything, I would prefer to have an option to dial back the impact of the lens correction which seems (in this module for the Sony RX100) to be a little aggressive.

From the original Fujifilm RAF file
From DNG after processing in DXO Deep Prime

If you don’t use Fujifilm cameras with X-Trans sensors, or if you have a lot of images shot on older less capable sensors, then DXO PureRAW might be worth it for you, despite the price. For people like me, who mostly shoot Fujifilm X-Trans, the calculation is different. Do I have enough noisy images from older non-Fuji sensors I want to clean up? Do I shoot enough with non-Fujifilm cameras? I like the program but I’m not quite convinced I need it.

Finally, here is the first image after processing in Deep Prime and some further processing in Lightroom.

Light and shadow

On the corner of Tsar Osvoboditel and Vasil Levski the entrance to the Sofia University metro leads to an open air concourse before disappearing from view under the road. From ground level a wide angle lens results in a nicely layered image. The light and shadow of the concourse leads into the bright band of sun illuminated cladding of the station wall and then the curved facade of Sofia University, topped off by a spring blue sky.

Walking past a few days ago I was drawn to the hard contrast between the sunlight reflecting off the light coloured stone of the concourse and the dark shadows of the interior. This is another one of those situations where, having found a good spot to shoot from, it’s good to hang around for a while watching scenes unfold. So after taking the initial wide angle shot I waited by the same spot for a while watching people come and go and trying to catch them as they moved between the darkness and the light, or were illuminated by the sun against the background shadows. Sometimes it’s a waste of 30 minutes, often you will get a few decent shots, occasionally you might come away with a gem. No gems this time, but I did get a few decent shots.

As with most of my recent photography I used my Panasonic LX5 with its small and, in digital years, ancient sensor. So given the very extreme contrast of light and dark I was not at all confident about how they would turn out. So when I saw the pictures in Lightroom I was very pleased with the outcome. While the unprocessed RAW files often looked like the highlights were blown, one click of the auto tone button transformed most of the images, not only retrieving detail from the highlights but also pulling detail out of the shadows, as you can see from the comparison below.

Lightroom auto exposure adjustment – before and after

In fact, in most cases the camera performed so well in retaining detail in the shadows that I had to darken blacks and shadows to get the effect I wanted. Here are a few that I liked.