Focus-testing results 50/1.8 poor!

[tubs]

I said I'd post the results of my focus testing, so here they are. I'm very disappointed. Even a cheap lens should do what it claims. In my experience, this lens should be sold as a 50mm/5.6. If anyone has a copy of this lens that can focus accurately at large apertures then could they please let me know ?

The history of my experience with the nifty-fifty:

1. Out of the box, shot 50-100 shots of moving subjects at f1.8. Only 10% of shots are sharp, but these are incredibly sharp. With my other two lenses 90% of my shots are pin sharp. Suspect DOF so switch to stationary subjects.

2. Several dozen shots of stationary subjects. 10% of shots are sharp. Starting to worry. I'm sure its not camera shake, but...

3. Switch to tripod. 10-15 shots, focussed both near and far and with both AF and MF. Only one shot is perfectly sharp. A few are useable but soft. Around 10 are unusable.

4. Post to this list. Many people respond that it is a sharp lens that works well.

5. Read, and research. Find other owners not happy. Sure its a focussing issue.

6. Post to list again. Read about Tim's focus testing. Decide to conduct the test myself.

7. Lens fails focus testing dismally in both AF and MF. Go back to store and exchange lens. Store guy said he's had no other returns.

8. New lens also fails focus testing dismally.

I really want to know whether its worth chasing this pony. Before I ask for a refund, can anybody convince me that it might be worth trying more copies of the lens ?

Here's how the testing went: Set up 20D at working distance 83cm from subject. Subject is ruler placed at 45° to lens axis. Camera on tripod. Room well lit. Took many shots using both AF (centre point only) and MF. Always focussed on the number 150 on the ruler. For each AF I initially moved the focus to either very near or very far, in case that made a difference. A typical result is attached below.

For each image I carefully studied the sharp region and calculated the centre point. A result of 150 would represent a working focus. I would accept any result that is within the DOF, so for f/1.8 anything between 135 and 165 would suffice.

Results - first lens - f1.8 - AF
125, 126, 111, 119, 85, 98
average focus error -39mmxcos45 = 28mm in front.

Results - first lens - f2.8 - AF
110, 130, 130, 145, 140, 128
average focus error -21mmxcos45 = 15mm in front.

Results - first lens - f5.6 - AF
155, 160 - hard to pick centre exactly
conclude accurate focussing at this aperture

Upshot: AF is both more accurate, and less important, at higher f/#.

Results - second lens - f1.8 - AF
102, 97, 124, 113, 112, 125
average focus error -38mmxcos45 = 27mm in front (same as first lens)

Results - second lens - f1.8 - AF - pressing AF twice before each shot
123, 125, 126
average focus error -25mmxcos45 = 18mm in front.
Marginally better and also more consistent than only one press. Still not good enough.

Manual focus results: All f1.8 and focussing to make the number 150 as sharp as possible.

First lens
140, 170, 132, 107, 117, 140
average focus error -22mmxcos45 = 16mm in front.

Second lens
118, 100, 118
average focus error -38mmxcos45 = 27mm in front.

I played with the MF for a while and I estimate my MF human erorr to be at absolute most +/- 15mm on the ruler, which is +/- 11mm straight, but it is most likely better than that. The upshot: You *dont* get what you see in the viewfinder.


Arrrrrrrrrggggggggggghhhhhhhhhhhhhh!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!

[tim]

That test isn't reliable in my experience, do this test (http://www.mrwild.co.nz/ExperimentalGallery/Tests/FocusTests/Set2/index.html) instead. I got completely different results when I did it that way. Focus on something parallel to the sensor, and put a ruler underneath for scale. Make sure you line the ruler up so a known point on the ruler is parallel with the box. See my results in this thread (http://photography-on-the.net/forum/showthread.php?t=66959&page=4).

[tubs]

Tim, do you have any source of information on why my test would be unreliable? I can't think of any reason the 20D wouldn't be able to AF off an angled surface. The only possibility would be polarisation by reflection, but the light from the object is diffuse, not specular, so I can't see that being an issue. In any case, my MF results agree with the AF ones, so I really can't see how there could have been a problem. Any thoughts ?

[tim]

Tim, do you have any source of information on why my test would be unreliable? I can't think of any reason the 20D wouldn't be able to AF off an angled surface. The only possibility would be polarisation by reflection, but the light from the object is diffuse, not specular, so I can't see that being an issue. In any case, my MF results agree with the AF ones, so I really can't see how there could have been a problem. Any thoughts ?

Nope, just my experience. Have a read of page 3, I think, of that thread I linked to. Hopefully Ron will find the article he was talking about.

[tim]

Don't take my word for it, try both types of test yourself. Also, posting loads of numbers isn't so helpful as posting 100% crops that contain full exif data.

[ed2day]

I've never bought into the argument that you HAVE to focus on a plane perpindicular to the camera. But I'm willing to be proven wrong. However I have found I get unpredictable results if I don't focus on a single line with nothing else near the focus point. So I don't trust "ruler tests". Look for one of the focus patterns with a single focus line in the middle and scales on both sides. The reason is you don't know what exactly the camera focused on. Don't assume it's what YOU focused on. The camera has a focus point that's larger than what shows in the viewfinder, about twice as large. So it may pick a point slightly outside the red square. And indeed I think it's programmed to lean towards a point that's closer to you if given a choice. I also found the long list of numbers too tedious too read through. Pics are better.

[Tom W]

Tim, do you have any source of information on why my test would be unreliable? I can't think of any reason the 20D wouldn't be able to AF off an angled surface. The only possibility would be polarisation by reflection, but the light from the object is diffuse, not specular, so I can't see that being an issue. In any case, my MF results agree with the AF ones, so I really can't see how there could have been a problem. Any thoughts ?

It can focus on an angled surface, but the problem is, the actual focus sensor is larger than the angled object and will obtain focus on any reasonably contrasty boundary that falls under the actual focus sensor's area of coverage. With the center cross sensor, this could be completely outside of the red focus square and above, below, to the left, or right of the indicator. On an angled object, that means that the focus can be achieved on any point along that object from about the width of the red square in front of - to the width of the red square behind the indicating square.

By having your focus target square to the lens, the entire focus sensor is equal distance from the object on which you are focusing and will, if working correctly, focus exactly on that plane (or within the tolerance accepted by the manufacturer) regardless of what part of the focus sensor detects the object.

[Tom W]

I've never bought into the argument that you HAVE to focus on a plane perpindicular to the camera. But I'm willing to be proven wrong. However I have found I get unpredictable results if I don't focus on a single line with nothing else near the focus point. So I don't trust "ruler tests". Look for one of the focus patterns with a single focus line in the middle and scales on both sides. The reason is you don't know what exactly the camera focused on. Don't assume it's what YOU focused on. The camera has a focus point that's larger than what shows in the viewfinder, about twice as large. So it may pick a point slightly outside the red square. And indeed I think it's programmed to lean towards a point that's closer to you if given a choice. I also found the long list of numbers too tedious too read through. Pics are better.

I agree that a planar object isn't necessary, but a single object of reasonable contrast with nothing else on which to focus is necessary. Even the single-line scales are a bit difficult as the camera can occasionally focus on something as simple as a crease in the paper. The strongest contrast point isn't always the one chosen by the camera to achieve focus. It sometimes will focus on whatever it "sees" first, whether it is the intended point or not.

[ed2day]

Tom,
Good point. One thing's for sure...these tests aren't trivial to do properly.

[tim]

Have a look at my tests (http://www.mrwild.co.nz/ExperimentalGallery/Tests/FocusTests/index.html) to see how different the results come out. The "set1-ignore" folder focuses on a slaneted scale, the "set2" folder focuses on a box that's parallel with the sensor. Set1 indicated that the camera was at fault, set2 indicated it was just my 2 50mm lenses.

[JaertX]

It appears either my EF 50mm f/1.8 II or my 20D is front-focusing. I did a very unscientific test based on tim's. I will certainly have to redo it. I need to clean my sensor as well...or so it appears in the small aperture pics.


Subject was approximately .75 meters away, focused on the lens box. Focus is clealy on the "4" instead of the "5" as it should be. Even the letters themselves are pretty soft!

Dad gum it!

[tim]

Is the 5 parallel with the box? It looks from that photo like the 5 is behind the box and that the 4 is what should be and is in focus - but the photo could be misleading. I'd also find a box that has writing very close to the edge and focus on that, DOF is very narrow and that could make a small difference.

As an aside, see how the photo's underexposed? That's because the scene is mostly white and the camera makes things around 18% grey. 1.5 or so stops of exposure compensation would expose it well.

[ron chappel]

Tom W describes the issues very well.
While i couldn't find the canon link i was looking for,i did find a very similar test for the D70 -which has the same tech difficulties for setting up a test
This article describes everything perfectly -with pics.
It also has a printable test chart that has the measurements offset so that they are accurate when tilted at 45deg :)

http://md.co.za/d70/chart.html

[JaertX]

Is the 5 parallel with the box? It looks from that photo like the 5 is behind the box and that the 4 is what should be and is in focus - but the photo could be misleading. I'd also find a box that has writing very close to the edge and focus on that, DOF is very narrow and that could make a small difference.

As an aside, see how the photo's underexposed? That's because the scene is mostly white and the camera makes things around 18% grey. 1.5 or so stops of exposure compensation would expose it well.

The 5 was as close to parallel with the box as I could get it on a fast and dirty set up. I didn't have a ruler, so I had to use an old achitectural scale that I still had. If you don't know, they are triangular, so I think that's what is throwing the perspective off.

But the problem is, the text on the box is not in focus either, so I'm not sure.

I definately need to redo this and take one or two shots with manual focus. I'll also print up the test chart that ron linked to and see how that goes. Probably a camera issue and not a lens issue though.

thanks!

[tubs]

Ok, I've read Ron's advice and read the article (http://md.co.za/d70/chart.html), printed out the test page and run the test again. As I suspected, results are identical to my previous test. (I guess my focus square was far enough away from the other features on the ruler in the first test.) After the clear advice not to post numbers again, I wont, but the upshot is that my second 50/1.8 front focusses by between 10mm and 30mm each time, regardless of whether I'm using AF or MF. The DOF is around 15mm (depends on working distance of course), so this explains why most of my shots with this lens are unusable.

The store seems happy for me to keep exchanging the lens until I find one that works, but its quite out of my way to go there. I'm desparing that there won't ever be a copy I can use. Can anybody tell me if there is an alternative brand lens for a similar price which might work ?

I'll post a link to one of the new test images tomorrow after I shrink one for the web.

[ron chappel]

Nice to see you've nailed the problem (or at least diagnosing it anyways:))

You should be able to do a test while you're there -i'm assuming a focus test should show easily on the review screen when at maximum magnification(?)
You could do it at a local park ,etc near the store.That way you wouldn't have to keep driving all that distance to exchange lenses.

[Bill Ng]

Tubs, I replied to your PM. On this thread though, what I don't understand is how the manual focusing got screwed up. I mean, what you see through the viewfinder IS in fact what you get. If autofocus was failing, you'd be able to see it in the viewfinder and compensate by manually focusing.

Anyone else have any ideas as to why it still ends up "off" even after manually focusing?

Bill in Brooklyn

[tim]

I mean, what you see through the viewfinder IS in fact what you get.

That's what I thought to, but with my copy of this lens if I manually focused you sometimes saw a different image on the PC from what was in the viewfinder. It wasn't a diopter thing either.

[ron chappel]

That sounds abit odd for sure.Your focus screen could be loose or even misscalibrated-but then all the other lenses would look out of focus through the viewfinder as well.

[tim]

Yeah, just the 50mm lenses front focusing and the rest not is confusing me. Canon want me to send the lens and camera back, but if I send the camera and they adjust it then maybe all my other lenses won't work. I don't want to send them all my lenses either, it'd cost a lot to ship securely and insure. I'm gona send them the 50mm lenses some time soon.

[tim]

Tom - in reply to your deleted post - the other lenses all work perfectly.

[Tom W]

Yeah, just the 50mm lenses front focusing and the rest not is confusing me. Canon want me to send the lens and camera back, but if I send the camera and they adjust it then maybe all my other lenses won't work. I don't want to send them all my lenses either, it'd cost a lot to ship securely and insure. I'm gona send them the 50mm lenses some time soon.

If all your other lenses are working right, there's no reason to send the camera unless they can guarantee that they will only calibrate the lenses to match the camera, and not touch the camera's calibration.

[Tom W]

Tom - in reply to your deleted post - the other lenses all work perfectly.

That's why I deleted it - spoke faster than my brain operated. :)

[tim]

If all your other lenses are working right, there's no reason to send the camera unless they can guarantee that they will only calibrate the lenses to match the camera, and not touch the camera's calibration.

That's what I figure too :)

That's why I deleted it - spoke faster than my brain operated. :)

I do that too ;)

[tubs]

Ok, after this I'm probably not going to push this barrow any more. I'm tired of giving myself an ulcer over a A$179 lens. I've received some wonderful advice from the group and for that I'm very grateful. I think its wrong that the lens doesn't focus at the wider apertures; I plan to try a few more copies of the lens and with luck I'll find one that's acceptable.

Some comments further to my previous ones:

Statistics are important for focus testing results, but I don't see the sense in posting dozens of test images, so I'll be posting a few numbers. These represent the front-focussing error in mm at f/1.8 and a working distance of about 70cm. One example test image can be seen at http://science-fs2.newcastle.edu.au/~cjrf622/. All done as per instructions for using that test sheet.

1. I can't explain why the MF doesn't work properly, but I believe it doesn't. In the test I manually focussed very carefully whilst watching the text 'This text should be perfectly in focus'. Resulting front-focus errors were (mm) +30, +20, +15, +5, +15. Its not a random error because then some of the numbers would be negative. I don't believe its a diopter issue because I have no trouble getting pin-sharp MF with my other lenses (I don't have any other 1.8 though.) I will note that the problem gets significantly worse with longer working distances. I've done innumerable testing with MF focussing on objects like billboards with large text. The only way I can get the text in focus is to find the viewfinder MF and then take a half-dozen images at successive tiny increments on the dial. One of them will be sharp even though the corresponding image in the viewfinder was blurry.

2. Someone suggested AI Servo mode. For the same test above, the results using AI Servo are +10, +15, +10, +15, +15. My observation here is that these numbers seem more consistent than the MF results (and the One Shot Mode results), so it seems the hunting does some good. My guess is that there is a systematic front-focus error of 10-15mm (at 70cm working dist), but for MF and AF One-Shot there is an additional random error. At this working distance the DOF is about +/- 10mm, so MF and AF One-shot are unacceptable 90% of the time, but I might declare AI Servo mode soft but acceptable at short working distance.

Ok, one more test - I'll go and try AI Servo at a longer working distance.

[tubs]

Just to clarify, the image I posted in my comments above was from a previous test where the working distance was a little larger, so the DOF is more like +/- 20mm, rather than the +/- 10mm in the results I quoted.

[RAitch]

I can't see why people first responses to posts like this is about AF tests not being accurate... especially when you complain that MF is failing.

"I played with the MF for a while and I estimate my MF human erorr to be at absolute most +/- 15mm on the ruler, which is +/- 11mm straight, but it is most likely better than that. The upshot: You *dont* get what you see in the viewfinder."
- tubs

I found the same problem and it drove me nuts to the point where I took the 50 out of my camera bag. It's a hunk of crap to me wrapped in plastic if you get a different picture than what you see in the viewfinder. If everyone's first reactions were true ("but try your test a different way because AF works different... blah blah blah") we'd all be better off taping over the viewfinder, pointing our cameras blindly at something, and pressing the release. The chances of getting a descent picture are probably better anyway.

So before anybody else starts going into reasons why AF may select points that are slightly different than the center of the focus point (because the focus zones are actually bigger), I ask, why would this then impact the image seen in the viewfinder? Are the people that find these lenses working perfectly just using the viewfinder for composition and trusting that AF will focus for them? I like to trust my own eyes. That, and obviously AF doesn't work either... but those are for the reasons that don't explain why MF gets weird results.

Refer to my original post if you haven't.
http://www.photography-on-the.net/forum/showthread.php?t=66835

[blinking8s]

ive broken 3 50mm 1.8's and one two of them the AF just got worse and worse until nothing was ever in focus...the other the glue messed up and the main front of the lens fell off after taking minor impact...