@ KeithR :
I know this blog, I already pointed it out. Even bettter, the blog post you've pointed at shows clearly that on a 7D, the photographer has all his viewfinder's marks fully covered by the AF sensor. Look at focus points 6 to 11, in my case those are only half covered. I've tested the same thing on two 70D bodies, and even some pictures you've posted (grass strand in front of a road) point toward a similar "particuliarity" in your camera.
Here's some pictures showing how this can affect you :
I pictured a lightpole against some foliage (not the best subject but had nothing else laying around). The lightpole edge is the most contrasted element in the frame, no reason for the camera to focus elsewhere. I used a tripod, with cable release and turned off image stabilization.
There are four pictures :
- In the left column, the pictures I took with the 70D. In the right column what I see looking through the viewfinder.
- In the top row, is the case where the lightpole is in the naked side of the focus point : the camera focusses on the foliage. In the bottom row, is the case where the lightpole is in the other half of mark : the focus is on the pole.
In my case, the 70D is the first camera i've come across with such "particularity". I'm calling this a "particularity" because otherwise the focus now seems to work well on my new 70D.
Don't get me wrong, I'm not criticizing Canon for having some overlap between each focus point, or for the focus sensitive area extending beyond the mark. What I'm criticizing Canon for, is that if they put in the 70D a different AF sensor than the 7D, they should have at least redesigned their focus screen marks to better take into acount the differences. I rely on those marks to frame my picures ! Now each time I take a picture I need to remember my graphic.
@cale_kat
I used these beautifull colors so you could easely differentiate each focus points, because some of them are overlapping. Other than that, the different colors don't mean anything.
For the data, I'm having trouble finding a way to send to you 334Mo of data (or more than 1Go if you want the RAW files), do you have any easy solution where I don't need to signin on some website ?
As for the legend, it was allready above the picture, but let me make it clear for you :
LEGEND :
I overlaid several pictures on top of each other showing the sensitive areas for each 19 focus points measured on my 70D.
- In black these are the focus marks you see in liveview AF Quick mode or in Canon's DPP picture editing software (by pressing Alt+L)
- The red, green and blue rectangles are the different AF sensitive areas for each focus points, the colors are here to help differenciate each focus point.
@cale_kat
"I don't think you're giving the camera credit for being as "smart" as it is. The autofocus is ignoring the pole because it has found a much higher contrast image in the back (the bush) and it focuses on that. The pole is a predictable undesirable-element (shape) in your picture and as a color it lacks contrast, being almost uniformly grey."
I'm sorry but you don't understand how phase AF works, read this and understand it before saying anything else :
http://graphics.stanford.edu/courses/cs178/applets/autofocusPD.html
Regarding face detect, the phase AF simply can't detect faces, where did you learn this ? Face detect is only available in Liveview.
That's because phase AF only takes a small portion of the picture, to be more precise, the AF system only sees a line or two of the picture. I can only say that my "focus sensitive area" term is incorrect, because in reality I should have drawn a cross and not a rectangle. But in no way can the camera guess from a simple profile plot where a face can be. Also, the metering system hasn't got enough resolution to detect faces, it is constituted of 63 pixels.
So phase AF system can't make any face recognition. In other words, the camera is simply not intelligent enough to ignore the pole. The only intelligent thing the camera can make is to track a moving subject. That "intelligence" must come from the fact that a tracked moving subject will keep a similar contrast pattern in regard to a moving background contrast pattern. But in the picture I presented, I wasn't tracking a light pole, I was on a tripod, with ONE SHOT enabled.
But maybe you can provide me a link of a Canon engineer saying that the 70D can do face detect while aiming through the viewfinder...
Also in my post, I spoke of the LIGHTPOLE'S EDGE ! The edge of the pole is the most contrasted element in the picture and, photogenic or not, that will be used by the camera to focus. How can't that black/white edge be less contrasted than the background ? I ploted below an horizontal grayscale profile plot across the entire picture. DO UNDERSTAND that this plot is across the entire picture, and not on the small area covered by the AF system. The light pole is the one with the highest intensity value between 2000 and 2500 pixels. That the kind of data the AF works with.
In conclusion, what I'm trying to point out is that the AF system simply can't see the pole's edge until it is well inside the viewfinder's mark. It's not a problem of the camera being smart or not.
"Honestly, Molybdo42, if you're going to go so far as to make a graphic which "proves" a point, you ought to be prepared to back in up. Explaining that your data is so raw as to be un-presentable in its present form is not proof but a failure to acknowledge that evidence can only be offered if it is understandable and accessible. Your presentation fails to qualify as evidence of anything."
I finally managed to find a way to post it (not because you helped). I took me hours to upload them but here you go, the raw data as you asked :
http://wikisend.com/download/316742/40mm_f2-8 RAW.zip.001
http://wikisend.com/download/152468/40mm_f2-8 RAW.zip.002
http://wikisend.com/download/962854/40mm_f2-8 RAW.zip.003
http://wikisend.com/download/288176/40mm_f2-8 RAW.zip.004
I had to split the archive in four, download the files (there's a 7 days time limit); put the files in a folder and follow these instructions to uncompress them :
http://www.techulator.com/resources/6601-How-split-join-files-using-7-zip-free-tool.aspx
I don't ask for anything more than any other person asking for a complete description of something you say is scientific but which fails to meet scientific, even casual science, standards.
If you were that "scientific" in your methodology, you would have already done the same test as I did and posted a similar picture.
BTW, the description of my measurement method can be found here :
http://photography-on-the.net/forum/showpost.php?p=17046799&postcount=1088
I then used a MATLAB code to merge the different pictures :
clc;
clear all;
NameFichier='_MG_94';
extImg = '.jpg';
threshold = 0.3;
for i=1:4:4
imageResult=NaN;
for j=i:i+3imageData = imread(strcat(NameFichier, num2str(j), extImg));end
imageDataBW = im2bw(imageData, threshold); % turns image in black and white
if isnan(imageResult)imageResult = imageDataBW;elseimageResult = and(imageResult,imageDataBW);end
imwrite(imageResult,strcat('Merge',NameFichier, num2str(j), extImg));
end
