The next release of G’MIC (numbered 1.6.9) is planed for the next week. Today, I’ve released some pre-release binaries for Windows and Linux (32 and 64bits). The git repo is also up-to-date. Is there any volunteer to install this pre-release version and tell me if all is working as expected ?
There have been a lot of improvements in the embedded math evaluator, and I might have missed some (probably minor) regressions.
Your help to hunt those remaining bugs (if any) will be highly appreciated
Here is the link where you can download and install the pre-release version of G’MIC : Index of /files/prerelease
Just saw your request at GIMPChat, David. Tested a few presets (GIMP plugin only) and all appears well (my usual ones and also the Continuous Droste preset that uses the new math operators). Looking forward to the new release next week. Will continue to use the pre-release until then.
Hello, found a little glitch: Various > Custom code (global) – no preview image (Local is okay). And both Custom code windows show a button with the Catalan text Actualització instead of Update/Refresh.
For me, despite my bad English, it isn’t necessary to have G’MIC translated in Dutch. Almost all tutorials for GIMP and G’MIC are in English. So I think it’s handy to keep it English, otherwise I have to translate the Dutch words back to English to understand the tutorials. But I’m not everyone of course. Other people might benefit from a translation.
I know of just one person who wrote tutorials for G’MIC in Dutch. He is already here. Thank you Paul, I am still waiting for your tutorial on the G’MIC color grading filter.
Hello, I have been looking a moment at the amount of work a translation of G’MIC would generate. I did that for a couple of years for RawTherapee btw. But the problem is to keep the translations updated, I know, I know!
But I thought that someone who’s using G’MIC in GIMP does not really need a translation. A certain general understanding of how filters work (and plain trying) is enough to get started.
Iarga, thanks for the compliment, I stopped for a moment with my Dutch blog about GIMP and G’MIC (among others) in order to set up a new English site with a broader scope (but still photography oriented). And for the moment I’m busy with the new newwavelets branch of RawTherapee, people are also asking for tutorials. But before one can write those, one first has to understand what wavelets are about and that’s a bit complicated, to say the least!
Anyway, seen the number of visitors that come since more than a year to my Dutch site (not millions a day nor thousands, but enough to see there’s an interest), I think it is a good idea to continue there as well. So looking for a way to create bilangual articles in one sweep in fact! I promise I’ll write an article about the color grading filter, as it is interesting enough!
The real problem is that there are too many very interesting things going on in this open source world, hard to keep concentrated on one subject!.
This is perhaps also the same with the “Samj corline b” filter (very nice filter). In the preview it shows much more contrast then after after “OK”. So I can’t trust the preview now.
I tested “Color balance” with a lot of photos. In the preview you can see difference with stretching and without stretching. Then after “OK” it displays something else then in preview. The strange thing is that this is not always the same. With different images there are different results. Some photos will stretch a little, but there are images that will fade even more and with some you see no difference after “OK”, while in the preview there was a difference.
So please use a lot of different photos. Compare “Color balance” with and without stretching. Observe that it behaves freakish.
Hello, yes I confirm: previews in G’MIC are not always correct or accurate. (You preview your photos with preview size at Largest, I suppose?). So do not rely on previews, just apply a filter to the whole photo and if you are not happy with the result, do a Ctrl+Z, for Undo.
But I think this is a bug. Because the stretching does little to nothing on a lot of images. I remember that in the past, this was different. But I also think this problem existed before the last update. So please look at this.
Some things to know about computing a preview (in general, not only related to G’MIC).
The G’MIC plug-in uses the default GIMP preview widget to retrieve the image data to compute the preview. This widget can be basically used with two different zoom factors :
In mode 1:1, what you see in the preview is a crop of the (usually) bigger image, where the size of each pixel is preserved (image is not upscaled or downscaled). This mode is used to preview filters that often use image data on local neighborhood (e.g. 3x3 or 5x5) to render. In that case, the preview will be accurate only if the filter does not rely on more “global” image information (which would be the case if it needs a pixel value located in another position that is not visible in the crop). Typical filters that uses that preview zooming mode are the Smoothing or Sharpening filters.
In ‘global’ zoom mode, what you see is a downscaled version of the whole image. This mode is useful for filter that renders things which requires the whole image geometry, like Frames, Tileable rotations, etc… In that case, the preview will be accurate only for filters that does not rely too much on the computation of ‘local’ informations.
Of course, most of image filters (in general) use both “local” and “global” information, which means there is no ways to compute an accurate preview based on a crop or a downscaled version of an image. The only way to have an accurate preview for all filters would be to compute the filter on the full size image. This is what recent GEGL-based filters do with the preview-on-canvas feature, but of course this can be done only for very simple and fast-to-compute filters.
To me, a preview is just a preview. It does not guarantee you’ll get exactly the same result (sometimes it’s possible to ensure that, but it’s in very rare cases). I don’t think the on-canvas preview would be that much useful in the case of the G’MIC plug-in neither (otherwise, as Paul suggested, just apply the filter and Undo it if this is not working as expected).
That being said, it’s exactly what happen with filter ‘Color Balance’ in some case: The “Color Balance” filter basically retrieves some color statistics from your image, then do some transformation based on their statistics. The “stretch” feature normalizes the final image into [0,255] basically. The preview for this filter is using a “global” view mode of course, so that the global image statistics computed on the preview are as close as the statistics of the original image as possible. But close doesn’t mean same as.
Now imagine you have a big image with very few contrast in it, except on a few points where you have very dark or bright values (a bit like Salt&Pepper noise). The [min,max] of your original image will be for instance [0,255].
But what will be the [min,max] of your downscaled preview ? Probably something very different, due to the inner smoothing done by the subsampling, so maybe something like [60,100]. The stats will be very different between the preview and the big image, and the filter won’t show an accurate preview.
This is one example, but imagine all of the 450+ filters have to manage this kind of issues.
So, to make it clear again : There is no ways to compute an accurate preview (except for very basic filters) from either an image crop or a downscaled version of the image. This is not a G’MIC-specific problem. This is an obvious fact for all image processing filters.
Thank you for your detailed answer. I think I understand your explanation, The normalizing stretches everything including the noise. Because the noise is already on minimum and/or maximum, it stretches less or doesn’t stretch. I use G’MIC a lot for scanned old photos. Of course they are dusty and noisy, so they don’t stretch. Downsampling images (as in preview) makes them less noisy, so they stretch more.
Seeing is believing. I applied a Gaussian blur of 30 in GIMP on a old scanned photo. Thereafter I applied “Color balance” in G’MIC with and without stretching. Now the difference was very clear after “OK”. Again much thanks for explanation.
