https://www.dropbox.com/sh/1s82dhko4k1jqvr/AADH9Gb2qvoVi-5XL33PmqT3a?dl=0
"Classic" raw backend (unified, lua_fix, everything but crop_rec_4k):
Experimental raw backend (crop_rec_4k branch,
updates discussed here):
Just for fun - raw overlays on the processed image (as if the captured image were the output of cr2hdr) - with experimental raw backend from crop_rec_4k (this is a hypothetical scenario):
I've got these images by emulating the silent picture module on top of the 60D firmware, using the above CR2 as reference image. EOSM/700D/650D/100D were a little stubborn; the 60D has the same resolution, but different offsets, so the screenshot might be off by a few pixels. This method uses our own preview (so that one won't match exactly what you have seen on the screen, as the previews on regular image capture are rendered by Canon's image processor - a black box for us) but the raw overlays are the same as with a regular image capture (not yet emulated). So, I expect the raw overlays to be the same as on the real camera (maybe off by a few pixels because of different offsets).
Notice the solid areas in the last two runs (both with crop_rec_4k's raw backend) are overlapping very well - the solid zebras on the unprocessed dual iso image (what you'd get when capturing a dual iso image) correctly predicted the overexposure from the end result (after processing with cr2hdr).
The diagonal zebras are also present - most of image was rendered as half-resolution by cr2hdr (as only the lower ISO had useful data). The exceptions were the dark TV screen and some darker areas on the flower. This matches cr2hdr output for this image:
Semi-overexposed: 85.00%
Were the raw overlays different on the camera screen? Can you provide a screenshot?
edit: by playing around in zebra.c, I've got the individual zebras for each sub-image (low and high ISO):
Notice something that resembles the focus pixels in the high ISO image - there are traces of them in the above CR2
