[ QUOTE ]
While yes I have elaborated quite a bit on previous posts that I do a bit of resharpening when I sample down an image. What you are missing in your "test" is that more than likely the image will not be uploaded at 601 pixels on its longest side. It would more than likely be either 800, 1024 or 2048 (or larger) on its longest side. Thus if I resample down that large an image, it would really soften the image so I apply a small amount of resharpening. If you are uploading an image that is 601, that isn't a realistic test other than to prove what I have said a bunch of times already. If you want to do your tests again, then do it on an image that is 1024 or 800 pixes on its longest side and see the quality difference.
[/ QUOTE ]
Okay, I tested it with a 800 pixel size image. Here is the original image at 600 pixels that I resized on my own computer:
Here is the image that I uploaded that was the exact same image, but I had it sized at 800 pixels high when I uploaded it. I then copied it to my computer from the Portfolio:
As you can see there is a fantastic difference. The photo as resized by GG looks grossly oversharpened. You can see this around the eyes, the copyright notice, the beads, and the overall tonality of the photo. The resized photo is a very poor representation of the original. This is exactly what I see in hundreds of photos that are posted by people to the GG portfolio.
Now granted, if you are on a 1280 x 960 or higher resolution, you don't see nearly as many of the over sharpening artifacts. But since the majority of viewers are still on 1024 x 768, you really see a huge difference. So please view the above two photos on 1024 x 768 so you can get the full impact of the problem.
NOTE: One of the flaws in doing auto sharpening when you resize, is that most glamour photos are meant to be somewhat softer than normal images. So when you sharpen and image that has been softened it really makes the image look like the person posting it didn't know what they were doing. I still think a pop up warning would be a good stop-gap solution.
Cheers,
rfs