capture.hdf5 - Githubissues

diegoroyo / tal

(Your)-Transient Auxiliary Library - Toolkit for simulation and analysis of time-resolved light transport captures - pip install y-tal

https://pypi.org/project/y-tal/

GNU General Public License v3.0

19 stars 4 forks source link

capture.hdf5 #1

Open syjjsy opened 2 years ago

syjjsy commented 2 years ago

Hi， Cool work！

I hope you can provide a sample file to test the correctness of the installation package. Thanks！

diegoroyo commented 2 years ago

I have uploaded a sample file. There you can find:

testz.hdf5: sample capture file
testz.yaml and Z.obj: files used to generate the sample capture file

A small check you can do is tal plot xy_interactive testz.hdf5. If you move the slider to t=1689 or nearby, you should get something like the image below.

From there you can use the documentation in the README to check other functions. Do not hesitate to ask if you need help :)

Akua1919 commented 1 year ago

Hi, I have 3 questions.

It is mentioned that the time bin in your setting is 0.003ps, is it reasonable or is it 0.003ns in fact?
When I render the scene with 'confocal' setting, it seems that there is an error "ValueError: could not broadcast input array from shape (4096,x,1) into shape (4096,)" caused in tal/render/render.py around 526 line.
Does it support running on gpu? Mainly I mean if mitsuba can run on gpu? If yes, how to run on gpu?

diegoroyo commented 1 year ago

Thanks for your interest :) I'll answer in the same order:

The time bin (temporal resolution) is set by the bin_width_opl in the YAML config file. It is measured in optical path length or OPL (i.e. light travel distance * medium refractive index). Thus, it is measured in meters, and a value of 0.003 (3mm) corresponds to 0.003 m / c = 10 ps, with c being the speed of light. I think the time reported by tal plot xy_interactive (...) (5.067 ps in the picture above) is incorrect, it will be fixed in a future update.
Thanks for the error report. These two changes (commit 1, commit 2) should have fixed it, I believe. Re-build mitsuba2-transient-nlos, and pip install -U y-tal to version 0.7.1 to use the new changes. If you have more issues I'll be happy to help.
Most of the tests have been done on CPU. Testing on GPU is limited, if you want to run on GPU you will most likely have to tweak the code. Rendering with high resolution can take some time.

diegoroyo commented 5 months ago

Hi,

Bumping this issue to let users know that we released our implementation of transient and NLOS rendering in Mitsuba 3, which affects issues in this thread.

Now we fully support vectorized CPU and GPU execution thanks to Mitsuba 3's Dr.JIT, with a much faster implementation. Check it out if you're interested: https://github.com/diegoroyo/mitsuba3-transient-nlos