Jump to content
Our response time is longer than usual currently. We're working to answer users as quickly as possible and thank you for your continued patience.

ch22

Members
  • Posts

    75
  • Joined

  • Last visited

Profile Information

  • Gender
    Not Telling
  • Location
    France

Recent Profile Visitors

782 profile views
  1. Sorry for the delay. I can no longer upload to your dropbox, but if you are still interested, you can get the file at http://www.oitregor.com/divers/boxblur_divide_bug.afphoto. The bug does no longer appear with the 10.1 version directly downloaded from Serif website... but I met up against another one ( ) boxblur_divide_bug.afphoto
  2. I met up against a spectacular bug when applying a Box Blur live filter (in Divide blending mode) to a RVB pixel layer. I had discovered this bug in version 1.10.0 (MacOS and Windows), but soon after I tried beta version 1.10.0.251 and found that the bug was gone. So I said to myself that it was already fixed and that I just had to wait for the next official version. But 1.10.1 came, I tried it, and I saw the bug was still there —at least with MacOS. I tried to see if the topic had already been discussed but I couldn't find anything so I am sending this message. I had prepared a detailed report, but I left it in French (sorry!). Things are a bit more complicated than the previous presentation, because this bug combines with a display artifact that I think is inevitable when the scale factor is not 100%. The bug itself apparently affects all blur filters, but much less dramatically than the Box Blur. Au départ, j’ai voulu reprendre une vieille recette pour transformer une photo en dessin au crayon, à savoir dupliquer la photo, inverser la copie, la mettre dans le mode de fusion Densité Couleur – (l’affichage devient tout blanc) et appliquer un flou gaussien à l’une de des deux copies. Généralement, on passe ensuite en N&B en désaturant complètement et on obtient une version au crayon de l’image de départ ; il ne reste plus qu’à ajuster le rayon de flou pour obtenir le rendu désiré. En principe, on devrait obtenir le même résultat en désactivant le réglage d’inversion et en remplaçant le mode de fusion Densité Couleur – par le mode Diviser. Ce n’est apparemment pas le cas, et en plus le résultat change quand on zoome sur l’image, ce qui fait deux effets désagréables. Pour bien voir ce dernier effet, on peut fusionner toute l’image dans un nouveau calque, et là, surprise, on n’a pas le même affichage, sauf si on zoome à 100 %. En fait, je ne suis pas trop surpris de cette dépendance avec le facteur d’échelle car je l’ai déjà rencontrée plusieurs fois. Je l’attribue à un artefact de la construction de l’affichage, une sorte de moirage, quand l’image contient plusieurs calques avec différents modes de fusion. Mon hypothèse est que ce qu’on voit sur l’écran ne résulterait pas d’un calcul complet de l’image avant de la réduire aux dimensions de l’écran, mais qu’on ferait cette réduction calque par calque avant de prendre en compte les modes de fusion ; autrement dit, il n’y a qu’à 100% qu’on voit réellement ce qu’il y a dans l’image, et si on veut l’examiner avec un facteur d’échelle plus petit, il faut tout fusionner dans un nouveau calque. Mais il y a autre chose. Même à 100% d’échelle, on n’obtient pas le même résultat qu’avec le mode Densité Couleur - du moins tant qu’on conserve le calque de flou. Il faut le fusionner, et là, on va bien retrouver la même chose, mais c’est l’impossibilité de voir l’image définitive pendant qu’on fait le réglage, et c’est aussi un renoncement aux méthodes non-destructives pour le travail sur les images. Il semble donc y avoir une sorte d’incompatibilité entre le mode Diviser et le calque de flou. Les choses vont devenir encore plus spectaculaires quand on passera du flou gaussien au box blur. Lorsque le rayon est très petit, on va passer par un vrai feu d’artifice sur les couleurs primaires et secondaires, totalement inattendu et passablement suspect, à nouveau avec un affichage dépendant du facteur d’échelle. Enfin, là encore, on ne retrouvera le résultat attendu, c.à.d. quelque chose de voisin du flou gaussien pour la même valeur du rayon, que si on fusionne le filtre avec son calque parent.
  3. Hello When enhancing a layer, one can check that identical results can be obtained by (i) Adding a high-pass filter in linear light blend mode (ii) Or adding an unsharp mask filter (in normal blend mode) with a gain factor of 1 and a radius curiously threefold larger than the high-pass radius (for instance, 10 px for the high-pass filter and 30 px for the unsharp mask) The same comparison performed with a well-known competitor software (somewhat more expensive) cannot lead to the same exact identity because of different definitions for the high-pass filter, but one can look for the best agreement when varying the gain and the radius in the unsharp mask. Then, the best radius is not exactly that of high-pass filter, but it remains close to it. I wonder whether the radius scale read in the unsharp mask dialog box of AP is not threefold too large. Obviously, this does not prevent the good operation of unsharp masking in AP
  4. I'm not sure how DR evaluated in the equation? This is not a real equation, but a kind of pseudo-code. DR is the previous DR in the right hand side and the new one in the left side. For instance, the product blend mode would be obtained with DR=DR*SR (to be understood as DR_new=DR_previous*SR). Only the right hand side is to be entered into the AP box roundup(0.5-0.3*SR-0.59*SB-0.11*SB) Should that be 0.59*SG? Of course. I suffered from many mistakes in the process, I corrected a lot of them directly in the AP box and I did not always reported these corrections in my text processor... and unfortunately I used this text version to prepare my post. Sorry! Actually, I did not dare to directly deal with color pictures. I began the work with grey pictures, where the code is far simpler. Not that simple to get it right the first time, but I succeeded fast enough to be reasonably confident to tackle the color case. Now, I want to emphasize that my misadventures are not at all a condemnation of this Apply Image menu. Simply, in case of complex equations, better to have a true text editor at hand with a compatible copy and paste function!
  5. First, thanks to dmstraker and Cecil for their encouragements. As for the English conversion... sorry, but that's another story ! Now, returning to the dmstraker's experiments with the Apply Image menu, I tried too. And incredibly —but VERY painfully— I succeeded in emulating the Contrast Negate blend mode. The issue is that equations to be entered are fairly long (see below) so that entering them by hand without errors is nearly impossible. You must prepare these equations on a separate text editor and then copy and paste them into the right box in Affinity Photo. Using word processors (such as Word or Text Edit) is a poor idea as the pasted formulas are not understood when they are too long — I suspect they are ruined by undesirable hidden formatting characters. Better to use a true text editor ; unfortunately, I use a very old one (AlphaX), and the copy and paste from it to AP does not run ! Eventually, I mixed parts of copy and paste with parts of manual typing, and finally I reached the holy Graal, I mean the identity between the blend mode and its emulation... The pseudo-code for the equations to be entered is as follows : DR=(1-DR) * SEL + DR * (1 – SEL) ; DG=(1-DG) * SEL + DG * (1 – SEL) ; DB=(1-DB) * SEL + DB * (1 – SEL) where SEL is the logical expression ((SL<0.5) AND (DL<0.5)) OR ((SL>0.5) AND (DL>0.5)) , where SL and DL stand for the luminosities (lumas) in the source and target layers, that is SL=0.3*SR+0.59*SB+0.11*SB and DL=0.3*DR+0.59*DB+0.11*DB. Then, as pointed by dmstraker, replace the conditional expressions such as (SL<0.5) or (SL>0.5) by numerical evaluations roundup(0.5-SL) or roundup(SL-0.5), then replace AND operator with a mere product , and OR with a mere addition (this is valid in this special case). We thus get DR= (1-DR) * ( roundup(0.5-0.3*SR-0.59*SB-0.11*SB)*roundup(0.5-0.3*DR-0.59*DB-0.11*DB) +roundup(0.3*SR+0.59*SB+0.11*SB -0.5)*roundup(0.3*DR+0.59*DB+0.11*DB -0.5)) + DR * (1 – ( roundup(0.5-0.3*SR-0.59*SB-0.11*SB)*roundup(0.5-0.3*DR-0.59*DB-0.11*DB) +roundup(0.3*SR+0.59*SB+0.11*SB -0.5)*roundup(0.3*DR+0.59*DB+0.11*DB -0.5))) and similarly for green and blue components. These are the real formulas to be entered into AP If you get to this point and AP is not grumbling about your formulas, congratulations! You should have reproduced the result of the Contrast Negate blend mode!
  6. Indeed, the macro results only in the invert adjustment through the selection about luma comparisons (this can be seen in the video). Its purpose was simply to give the same display as the investigated blend mode, without modifying the RGB components of the layer; additionally, the selection can be recovered for any other fanciful operation you could imagine on this layer. My current version is 1.7.3 Let me some time to consider your formulas —so far, I never walked through this menu in my exploration of Affinity Photo (my personal project consists in in-depth explanations of the software for French photographers through a set of video tutorials, french speaking, of course! The job is not ended. See http://www.oitregor.com/numeric/affinity_photo/contenu.html )
  7. Sorry, Dave, I cannot open your file with my current version (1.7.3) —I got the warning “The file includes features from a later version of Affinity“. As for the right operating process, I join a short video demo_contrast_negate.mp4
  8. By November 2017, dmstraker was asking for information about Contrast Negate (https://forum.affinity.serif.com/index.php?/topic/49811-your-big-friendly-guide-to-layer-blend-modes/&tab=comments#comment-250891) . As I found nothing about this point in more recent posts, I give my opinion. I think that the luminosities of the two layers are analyzed, but in the sense of lumas, i.e. L'=0.3R'+0.59G'+0.11B' where all quantities are normalized quantities varying in the (0,1) interval. First a selection is built of areas where lumas are both below 0.5 ou both above 0.5. Then, the upper layer is replaced with its negative within the selection. I enclose a macro, simu_Contrast_Negate.afmacro which (hopefully) does this job —to be imported in the Macro panel. Select the upper layer and run it. An Invert adjustment should be created above the upper layer (allowing to recover the selection from the adjustment mask) simu_Contrast_Negate.afmacro
  9. Another method for obtaining the saturation mask or the equivalent saturation selection, which requires no HSV capacity : (1) open an HSL adjustment, put saturation to -100 and put the blending mode to "difference" (2) open a level or curve adjustment so as to get a x2 gain (for instance, with levels, put the white glider at 50%). This amplifying step will allow to get a full saturation mask, i.e. with white or black corresponding to fully saturated colors or gray tones. At this stage, the (max, med, min) triplet for color components is replaced with (max-min, xxx, max-min) where xxx is lower than max-min (3) in order to retain only the difference (max-min), open the channel panel and successively (i) right-click the red line for " load in the pixel selection", (ii) and (iii) right-click the green and the blue lines for "add to the pixel selection" —this is equivalent to choosing the highest of the three RGB components. We then have the wanted saturation selection.
  10. Additionally, switching to English in Preferences dialog makes the Overlay item reappear in the drop-down list with seemingly the expected behavior.
  11. I confirm. The second occurrence of the "Superposition" (Screen) mode is not a mere mistake in the drop-down list names, it really calls for the Screen mode.
  12. Actually, since a RAW file does not contain any ICC profile, the luminosity and the colors of a RAW file just opened does not matter very much. It's only at the output of the raw software that a ICC profile is added to the file and that RVB components are given a true color meaning. However, the camera JPEG file or the display by the system commodities or Adobe software (and many others, of course) are in reasonable agreement and thus offer a convenient reference with which Affinity Photo should comply — and unfortunately it does not always. Under MacOS and with Serif RAW engine, I observed nothing special with NEF, CR2 or ORF files, but I actually got this odd darkening with RAF files. However, it disappeared when I switched to the Apple RAW engine. Below are compared screen copies for RAF and CR2 between AP 1.6, AP 1.7 and Photoshop (in this case, with no correction) and a screen copy with the Apple RAW engine :
  13. I'm just one year late, roughly, but I discovered this discussion only recently by chance and I'm grateful for the appreciation. Actually, I wrote these tutorials with true beginners in mind and I tried not to lose them when advancing in the intricacies of AP. Now, by the beginning of 2019, many of the tutorials which were missing by the time of the jmmermet post are now in line. Only for french speaking people, I'm sorry, nobody is perfect ! Let me repeat the adress http://www.oitregor.com/numeric/affinity_photo/contenu.html
  14. Under MacOS, I compared various prints of a gray chart on a matte paper, on the one hand from Photoshop with perceptual intent (without black point compensation), and on the other hand from AP, either perceptual or relative, with or without black point compensation. All of them are practically identical.
  15. Sorry, my reference to the Photoshop model was somewhat irrelevant... and it does not matter. The key point is, whatever the actual model used in the AP colorimetric model, that putting saturation to 0 in the HSL adjustment makes the (Max, Mid, Min) trio replaced with (Max, Max, Max). This can be checked in a practical way by monitoring what happens to RGB components of a color sample in the Infos panel. Then, switching to the difference blending mode makes the original trio replaced with (0, Max-Min, Max-Mid). Since the largest component is now Max-Min, the second HSL adjustment puts the three components to Max-Min.
×
×
  • Create New...

Important Information

Please note there is currently a delay in replying to some post. See pinned thread in the Questions forum. These are the Terms of Use you will be asked to agree to if you join the forum. | Privacy Policy | Guidelines | We have placed cookies on your device to help make this website better. You can adjust your cookie settings, otherwise we'll assume you're okay to continue.