Default effort level might be better at a lower value

[kliberty]

The program help says effort of -E 60 is the default, but some cursory tests I believe 10 - 14 may be a better default.

I attached the image I used, as well as a plot showing the size-png_size ratio and time-max_time ratio vs. the effort value. I tried a couple other images with similar results, though, I did not attempt to cover types of image data.

I don't mind running it for other images, our you can use the scripts I used.

[lehitoskin]

I'd like to see this kind of statistic run on a bunch of different kinds of images: raster, vector, etc.

In my mind, I view FLIF as a sort of PNG with better compression, which means I only use it on images with lots of solid colors. Photographs aren't quite what I'd expect to be encoded with FLIF, so showing statistics for only photographs isn't enough.

[kliberty]

I can do that. Would you consider the images in the benchmark repo to be representative? I might have to resize some of the smaller images to make the timing more accurate, though.

[kliberty]

Oops, I accidentally hit "Close and comment" instead of "Comment"

[lehitoskin]

Yeah, that might be fine. The more data that is available, the better, IMHO.

[jonsneyers]

I think -E30 is probably a better default.

[kliberty]

Why 30? I'm running all the images in flif-benchmarks/test_images right, but I can only let it run while I'm sleeping and since I tweaked it for better timing statistics it is taking longer than I expected..

So far I've completed :

• debian
• icip-corel
• Lukas-2D-8bit
• openstreetmap
• png-gallery
• pngimg.com-car-logos
• pngimg.com-hammers

[kliberty]

A website seemed like the best way to display the results. They're posted at https://kliberty.bitbucket.io/flif-bench/.

I few images a missing because of some unrelated computer issues...but there are still more than 900 tests available.