06:00 | Bertl | off to bed now ... have a good one everyone!
| |
06:00 | Bertl | changed nick to: Bertl_zZ
| |
11:49 | aombk2 | joined the channel | |
11:50 | aombk | left the channel | |
14:22 | Bertl_zZ | changed nick to: Bertl
| |
14:22 | Bertl | morning folks!
| |
14:22 | vup | morning :)
| |
16:29 | vup | Bertl: ok, so I looked at some more things
| |
16:29 | vup | for one, the "weird" pixels seem to always be the same between different exposures
| |
16:29 | vup | (I assumed that previously, but did not actually check)
| |
16:30 | vup | furthermore, for the odd rows, they are pretty much 50% of the pixels
| |
16:30 | vup | (so 25% in total)
| |
16:30 | vup | and they follow a pretty strong pattern
| |
16:30 | vup | https://files.niemo.de/weird_pixels_even_rows_even_cols.png
| |
16:30 | vup | https://files.niemo.de/weird_pixels_even_rows_odd_cols.png
| |
16:30 | vup | https://files.niemo.de/weird_pixels_odd_rows_even_cols.png
| |
16:30 | vup | https://files.niemo.de/weird_pixels_odd_rows_odd_cols.png
| |
16:31 | vup | white pixels are the "weird" ones, black are ones with approx linear response curve
| |
16:32 | vup | btw these horizonatl "stripes", where the pattern is the same for every row are 240 pixels high
| |
16:32 | vup | any clue why that could be?
| |
16:32 | vup | also anuejn, se6astian^
| |
16:38 | Bertl | so for about 240 readout rows the 'pattern' stays the same
| |
16:38 | vup | yeah
| |
16:39 | Bertl | interesting
| |
16:39 | vup | (240 readouts of two rows obviously)
| |
16:39 | vup | would be interesting if any changes if one does one-sided readout
| |
16:41 | Bertl | well, if the pattern would be uniform, I'd say some of the ADCs have nonlinearity issues
| |
16:41 | vup | uniform over all rows?
| |
16:41 | vup | yeah
| |
16:41 | Bertl | (still might be the case for the 'white' columns)
| |
16:42 | Bertl | but the shift definitely looks weird
| |
16:42 | vup | yep
| |
16:42 | Bertl | to be honest, I'd double check the evaluation code ;)
| |
16:43 | vup | oh I already did a lot
| |
16:43 | vup | but this pattern is even visible in darkframes for example
| |
16:43 | vup | (this pattern of every 240 readout rows being approx the same)
| |
16:43 | vup | also note, that this is super low light conditions
| |
16:43 | Bertl | from how many sensors do we have this data?
| |
16:43 | vup | one
| |
16:44 | Bertl | okay, so maybe we should check with a different sensor then?
| |
16:44 | vup | yes definitely
| |
16:44 | vup | (low light conditions -> 13ms exposure gives pixel values of 20 to 40 after subtracting the darkframe)
| |
16:47 | Bertl | and they were captured via HDMI?
| |
16:47 | vup | yeah
| |
16:47 | Bertl | but not at 30/60FPS I presume?
| |
16:48 | vup | why not? It should have been done at 60fps
| |
16:48 | vup | right se6astian ?
| |
16:49 | vup | I pushed the stuff to https://github.com/rroohhh/cmv12k_color_science
| |
16:50 | vup | flatfield_experiments.ipynb contains pretty ad-hoc / hacky code that I used for the last plots (and the hexbin histograms)
| |
16:53 | vup | but in principle the code is all super straight forward
| |
16:53 | vup | For me the biggest source of uncertainty is the actual capture process by the recorder
| |
16:53 | vup | but with the current code, the frame counters and the wrsel seem to check out
| |
16:53 | vup | and "normal" footage also looks correct
| |
16:54 | vup | so I am not sure what could be wrong there
| |
17:03 | se6astian | I captured at 24 fps so 48 fps hdmi
| |
17:04 | se6astian | What about creating a capture script to guarantee controlled conditions?
| |
17:05 | se6astian | And it can also including capturing a few snaps with axiom_snap
| |
17:05 | se6astian | Then you can compare hdmi to internal raw frames
| |
17:06 | vup | sure, that sounds pretty useful
| |
17:06 | vup | we probably need some kind of "script" sooner or later anyways
| |
17:06 | vup | maybe you can start one? Its pretty hard for me to do it without a beta
| |
17:07 | se6astian | Will do, getting back to Vienna tomorrow eveing
| |
17:07 | vup | Sure, no hurry
| |
17:26 | vup | (also the axiom_snap references sound like a really good idea)
| |
17:32 | se6astian | With full sensor register dump
| |
19:10 | vup | Sure
|