Just to clarify what is meant by AI Servo AF in these tests; it is not continuous shooting (burst) but the use of AI Servo AF to take one shot. The technique I used was to set the camera to AI Servo AF and then manually defocus the lens. Then I used the 2 sec shutter delay to count down and fire one shot. For the next shot I manually defocus in the opposite direction and repeat using the 2 sec shutter delay. This ensured the image was recorded without my hand touching the shutter button at the moment the shutter was actuated. I used the same method for recording the One Shot AF images. I also turned off the lens IS for these tests. I did some tests with AI Servo AF and continuous shooting (long burst) which showed similar consistency to the single shot method I have just described. I also ran some series using the Intervalometer set to 3 sec interval but of course there was no time to manually defocus between shots. I could have set the interval to a longer time and manually defocused in between shots and maybe I will give this a try next time.

Unfortunately not much bird photography at the coast over the Christmas and New Year period due to the weather, smoke, fire situation etc. However I have attached a couple of photos taken with the 90D to show the results at ISO 1600, in case people are interested. ISO 1600 results are excellent when well exposed, with good colour and very controllable noise. I've tried to pick a couple of pictures which demonstrate detail and sharpness, and cropped to show detail (rather than focusing on composition!). I haven't used ISO 3200 much but initial impressions are that the sensor is starting to struggle at this level although I think the results will be useable at a pinch with good post processing and a good quality noise reduction program, particularly if downsized for BLP purposes.

Grey Butcherbird photo is a 4500 by 3000 crop of original; details are ISO1600, 1/250s, f/5.6, 400mm
Wallaby photo is a 2400 by 2400 crop; ISO 1600, 1/800s, f/5.6, around 230mm.
(Both photos were processed in DxO Photolab 3.1, then downsized to normal BLP size requirements ie these pictures have been both cropped in raw and downsized during exporting.)

Simon

Good demonstration from Simon of the ability to resolve fine detail with the 90D. I am not surprised as in my testing it often resolves up to 140 cycles/mm on the resolution test chart. See attached image. The calibration for the horizontal and vertical 'chirped' resolution test patterns is 12 units = 120 cycles/mm, 14 units = 140 cycles/mm and so on. For comparison, the Nyquist frequency for the 90D sensor is 156 cycles/mm. The camera cannot resolve the Nyquist frequency because of the roll-off of the AA filter; no surprises. The resolution achieved by the 90D is the highest of any camera I have seen, again no surprises.

It is a wet morning here and I am home-bound so I thought I would share some of the results of my AF testing. I have had some inconsistent results with AI Servo AF and One Shot AF that I think may be related to ambient air temperature on different days when I did the AFMA tests. The Live View results are very consistent, as expected, so for now I have attached the statistics on the shot-to-shot variation for that case. See attached graph of the shot-to-shot frequency distribution for a sample of 75 shots. Yet again, no surprises, Live View focus is very consistent.

Thanks Ian. Incidentally I found the 90D is the first DSLR camera where I have felt the need to do lens microadjustment. Early shots were sometimes lacking in absolute crispness. After testing I concluded that the camera and lens combo (100-400 at 400mm) was slightly back focusing although was not always getting consistent results. However I have found dialling in a small adjustment (two units) to compensate has made the viewfinder focus just that little bit better, particularly for closer subjects where DOF is lower. Now shooting the same subject with viewfinder focus vs Live View focus, the results I am seeing are more consistent with each other.

I have now completed my testing of the 90D + 400 mm f/4 DO II AF system. The posting of final results has been delayed by some inconsistent measurements that have not been properly resolved. During the course of the work I made significant improvements to the AFMA measurement technique in an effort to minimize inconsistent results. The improved measurement accuracy has enabled me to easily observe the effects of one unit of AFMA adjustment and the impact on the shot-to-shot AF variance of lower light levels (higher ISO). All of the measurements in this study were made with a range to the target of 16 m and aperture f/4. Under these conditions the 90D + 400 mm f/4 DO II can easily resolve 100 cycles/mm and up to 140 cycles/mm. Using 100 cycles/mm as the minimum required for acceptable resolution, I was able to find the tolerance on defocus was about +/- 4 cm from the plane of best focus. In direct sunlight with ISO = 160, about 95% of shots were within this acceptable range. In cloudy and smoke affected conditions I made measurements with ISO = 500 and found the variance was about 3x greater than in direct sunlight. This confirms the folk-lore that AFMA adjustments should be made in good light to minimize the shot-to-shot defocus variation.

The method used to measure the relationship between defocus and AFMA was as follows. The camera intervalometer or 2 sec shutter delay was used to enable hands-off exposure. Between each exposure the lens was manually defocused alternately in a clockwise and anti-clockwise direction. Sixteen shots were taken for each AFMA setting; the resulting standard error was about half the change in the mean defocus for one AFMA unit. In addition to the results obtained in this way using phase detection AF, sixteen shots were taken using the contrast detection Live View shooting mode. The LV results were used to check for any systematic errors in the results for AI Servo AF and One Shot AF shooting modes. Results obtained on two different days are shown attached below. Ideally, the two sets of results should match but they don’t and this inconsistency from day-to-day has been troubling me. The measurement technique is beyond question so I have been looking to other possible causes of this inconsistency. It is known that the AFMA can vary with temperature and has often been reported by North American wildlife photographers using cameras with long lenses in winter conditions. I have been wondering if the inconsistency I am seeing in the AFMA calibration is due to changes in air temperature and the camera body exposed to direct sunlight. In this regard, I have used a white handkerchief to cover the camera body in an effort to prevent it from warming up when testing in direct sunlight.

The AFMA calibration curves consistently indicate that the camera is slightly back-focused without any AFMA adjustment. I have at various times determined that the right amount of AFMA adjustment should be -1 or -2 units but have settled on AFMA = 0. This results in a mean back-focus of about 0.8 cm as shown in the defocus frequency distribution shown below. With the aperture settings commonly used in the field, this small amount of back-focus is neither here nor there.

One final comment on the LV results presented earlier. The defocus variance when using LV is about one-third the variance of phase detection AF in good light. This is a significant difference and may have implications for the performance of mirrorless cameras. In theory, mirrorless cameras should have a lower defocus variance than conventional DSLRs and there is plenty of anecdotal evidence for this in practice.

Ian

I commend all the work you have done on this - very useful. Your results (together with other info online about the variability of 'traditional' DSLR focussing ) suggest the optimal technique should be as Glenn suggested in an earlier post - focus, shoot, repeat... - although it seems that generally we have depth of field to thank for minimising problems. With the 100-400, for example, the minimum aperture at 400mm is f/5.6 which will give reasonable DOF. It also suggests that with painstaking work such as macro using a tripod, it may be preferable to use live view if you wish to autofocus, or manual focus with the zoom-in function on the screen.

Having a back focus of nearly 1cm might be fine at 20 meters plus, but I presume it might be more problematic at 5 m?

Simon