Pcond -h results in very blurry image


#1

I created a 360 stereo over-under pano using Mark’s 3d360.cal script. For some strange reason, if I do pcond -h pano.hdr pano_h.hdr, the resulting pano_h.hdr becomes very blurry. For instance, this is the original:

and this is the pano_h.hdr:

What can be causing the blurriness?


#2

When luminance levels are very low in the image, pcond -h simulates the loss of acuity at scotopic levels. However, you shouldn’t be in this range with normal lamp levels. Are you sure of your photometry for this lamp? What does ximage or falsecolor tell you about the light levels you are simulating? Did you pass the image through something (like pcompos) that lost the exposure setting before sending it to pcond?


#3

That’s strange. The scene is lit by a gensky as well an .ies lamp and computer monitor. I don’t think it is this dark. Here’s a falsecolor -i scene.hdr > falsecolor.hdr of my regular perspective view, do these values look sensible?

If they look sensible, perhaps it is something that 3d360.cal is doing?


#4

Yes, the luminance values seem reasonable, so I’m wondering about your image processing pipeline. Can you post the output of “getinfo” on your pcond input?


#5

Sure thing, this is what I rendered it with:

rad -v 0 scene.rif OPT=saved.opt
X=2048; Y=2048; cnt $Y $X | rcalc -f cal/3d360.cal -e "XD=$X;YD=$Y;X=0.8087165355682373;Y=-1.2962751388549805;Z=0.4604237973690033;IPD=0.06;EX=0;EZ=0" | rtrace -n 12 -x $X -y $Y -fac @saved.opt scene.oct > out.hdr

… and this is the getinfo output.

$ getinfo out.hdr
out.hdr:
 #?RADIANCE
 oconv scene.rad
 rtrace -n 12 -x 12288 -y 12288 -fac -dp 4096 -ar 128 -ms 0.14 -ds .2 -dt .05 -dc .75 -dr 3 -ss 16 -st .01 -ab 3 -af scene.amb -aa .1 -ad 1536 -as 768 -av 10 10 10 -lr 12 -lw 1e-5 -av 0 0 0 -ds .01 -dj .8 -dt 0 -aE exclude.txt
 SOFTWARE= RADIANCE 5.1.0 NREL/googs 2017.08.21 (based on RADIANCE 5.1 Official Release by G. Ward)
 CAPDATE= 2019:01:05 21:08:03
 GMT= 2019:01:05 10:08:03
 FORMAT=32-bit_rle_rgbe

#6

Hmmm. Looks fine, but it may be the lack of a view or something in the way you’re processing the image with pcond that’s confusing things. Maybe you can post your HDR image(s) somewhere, along with the set of commands you are using to tone-map so I can have a look?


#7

Here’s the HDR image, and the command I am running is pcond -h out_small.hdr > out_small_h.hdr. To create out_small.hdr I ran it through pfilt -x /3 -y /3 to reduce the filesize because the original is 500MB.

https://share.dmca.gripe/mcge64D0FKn1iyum.hdr

It’s a temporary download link, so I hope it works for you. It might’ve changed the filename.


#8

It is the visual acuity simulation in pcond that is causing the problem. Since there’s no view stored in the image header, it assumes the smaller of the two image dimensions is 40° and the larger is scaled as if it is a perspective image. This results in an estimate of image acuity that is off by about a factor of 9 in your case, so it blurs pixels where it should not. Just switch it off with:

pcond -h -a input.hdr > output.hdr

#9

Just wanted to add that this utility (pcond and its nicely wrapped options set in the -h option) is to me a really great example of the power of Radiance over its raytracing brethren. The fact that a Radiance image is full of photometrically-accurate data that can be interrogated and tonemapped to tell such interesting stories, is a testament to Radiance’s awesomeness. The fact that this capability has been available over twenty years now, just blows my mind.