HAL0THANE  Everything Arras said earlier (below), I second. The only thing I can add is that I usually scan at 90 percent, and I have yet to see a false positive out of two scans of over 25'000 images each. As far as false negatives, I can't know, of course, but I can only hope those are as rare.

As Arras mentioned, though, it does take time. This is not unique to dupeGuru PE, though, as all duplicate image scanners will be time-bound. (For the mathematically inclined, it's O(n^2) -- each picture you add must be compared against all the others.)  
(Version 1.7.8)

DRAW  Still not bug free for my 25000 photos archive. For smaller archives I love this app.

Application Identifier: com.hardcoded_software.dupeguru_pe

Application Version: 1.7.6

Traceback (most recent call last):

File "hsutil/cocoa.pyc", line 47, in _async_run

File "dupeguru_pe/app_cocoa.pyc", line 179, in _do_load

File "dupeguru_pe/app_cocoa.pyc", line 118, in update

File "plistlib.pyc", line 102, in readPlistFromString

File "plistlib.pyc", line 77, in readPlist

File "plistlib.pyc", line 402, in parse

ExpatError: not well-formed (invalid token): line 591089, column 41

Relevant Console logs:  
(Version 1.7.6)

ARRAS  When it comes to finding duplicates in large image collections, there are two kinds of apps out there: simple apps that look for files with identical content, and more powerful tools that identify images that are slightly modified versions of one another. This app is one of the latter variety--a very powerful and effective dupe-finder that can find not only the exact copies of an image, but also resized versions, lightly cropped versions, and versions with labeling superimposed.

I've used a half-dozen other dupe-finding tools over the years, some running on Windows, some on Linux, some on MacOS, all on the same growing set of 15,000 images, and while each tool managed to find a few that the previous ones missed, dupeGuru PE identified 104 dupes that all the others had missed. Very impressive.

The one downside that comes with this level of thoroughness, though, is the time it can take to analyze large collections with any significant degree of fuzziness. The 15,000 images I threw at it took it about 45 minutes on an 8-core Mac Pro at 95% filter hardness and matching dimensions. That grew to 75 minutes when I told it to look for mismatched dimensions as well.

To use this tool effectively, you just need to know how to use the filter hardness slider, which determines how exact the matches need to be. If you just want a quick search for exact copies (i.e. what the simpler tools do), slide it up to 99% or 100% and you'll be done quickly. The default setting of 95% gets better results, but obviously takes longer. The further you reduce that setting, the more variation you allow in the matches, so you can find more extensively modified variations of an image that way, at the cost of more time.

In short, the effectiveness of this tool is fantastic, and if it were faster it would rate 5/5. Even as is, though, it rates a solid 4/5 once you learn how to use the filter hardness setting properly.  
(Version 1.5)

AKROBAT  I've used the Music Edition of this software and it works really well. The Photo Edition on the other hand was still busy on my Mac after 1 hour and judging by the progress bar would take perhaps another 3-4 hours. I have relatively few photos.

I tried it again on 2 directories containing 14 and 24 photos respectively and after 20 minutes it was not quite halfway.

No idea why the Music Edition is so good and this so bad.  
(Version 1.0.4)