I don't know why, but sometimes pointing from the center is not working and coordinates should be shifted. And i still don't know how to get right position in all cases.
Probably i should use cropPos and PanPos somehow.
Don't know what cropPos and PanPos is but I also found out that raw video isn't always centered in relation to the full sensor. Running exiftool on a CR2 file will give lots of useful information:
Sensor Width : 5280
Sensor Height : 3528
Sensor Left Border : 84
Sensor Top Border : 64
Sensor Right Border : 5267
Sensor Bottom Border : 3519
Unfortunately this doesn't make it into the MLV metadata and of course RAW has no metadata at all so we're left guessing how to center the pixel map.
Now that I can see what a dcraw ".badpixels" map looks like as an image file I searched around and collected several "dot removal" tools and made image files of their pixel maps. Note that other tools can also remove the focus pixels but some of them use chroma smoothing or other methods that don't require a pixel map.
Let's start with one of the earliest tools, foorgol's PinkDotRemover tool 650D
I was surprised by how "dirty" the image appears. According to the author he used a script to extract the position of the dots. It looks to me that many of the coordinates are mapping random noise. What seems strange to me is the pattern looks more like what I've seen on a 700D rather than a 650D yet the tool was made to work with 650D, EOSM and "Crop Mode" though I only found one pixel map file in the tool so it must be doing some sort of scaling. The pixel map covers a 1280x720 area.
Next up, maxotics unfinished FocusPixelFixer
Quite a difference. maxotics put a lot of effort into the "pink dot" issue and it appears that he has come up with a more orderly pattern. This is also 1280x720. I'm not sure if it covers both crop and non-crop but it seems that he has done most of his research using EOSM crop mode video so I would assume that this pattern is for crop mode.
The newest program that has dot removal, MLVProducer for Windows
AWPStar created not one but four different pixel maps. These are for two different focus pixel patterns in crop and non-crop mode. His pixel maps are different than the others because his x y coordinates are zeroed out at the center instead of the upper left corner like the others.
Let's look at his crop mode version, dot32_crop:
This is very similar to the one from the focuspixelfixer pattern but in a slightly larger 1792x1008 size. Note that the images I'm posting have been resized to fit into the forum's guidelines. If you want to examine the full sized images you can find them on my flickr photostream
The non-cropped version looks quite a bit different, dot32:
Compare this to the crop version number 2 which is different than the other three because this one I calculated to be 1280x720 while the others are 1792x1008. This one is also the closest in appearance to the one that I created for the EOSM crop mode: dot32_crop2
And the non-crop version number 2: dot32_2
Something that is interesting is the number of coordinates that were mapped.
|PinkDotRemover tool 650D:|
|MLVProducer for Windows dot32_crop|
|MLVProducer for Windows dot32|
|MLVProducer for Windows dot32_crop2|
|MLVProducer for Windows dot32_2|
My version for the 1280x720 EOSM crop mode has 6,120 mapped coordinates. So are more mapped coordinates better? I don't think so. I believe that only confirmed focus pixels should be mapped. I didn't use a script or a formulae to determine the position of each focus pixel, though I assumed that once I found a few points they would fall into a pattern. The most clear image that I found to verify my work was the one I posted on Reply #26
of this topic. I superimposed my pixel map with that image in Photoshop and at first things didn't line up very well until I found out that the original was a "FIT" file and those files have the 0,0 x,y coordinates on the lower left corner instead of the upper left corner like most graphic image files. Once I flipped the image things lined up much better. In fact I discovered that a few lines that I assumed were long rows were in fact short rows.
Now things are starting to make sense, at least for the
650D, EOSM I should be able to create a full frame pixel map file. I've still got to figure out how line skipping works in non-crop mode. Apparently it uses every third line but my question is, which ones are the third lines?
*[EDIT: Closer examination shows that the SL-1/100D doesn't match the EOSM/650D.]