## Search found 75 matches

Tue Aug 21, 2012 8:54 pm
Forum: General Development
Topic: MIS BPT with Russian Roulette
Replies: 16
Views: 13728

### Re: MIS BPT with Russian Roulette

Thank you apaffy. :) I tried to check uniform weighting (in this case, there are 3 strategies [s=0, t=4], [s=1, t=3], [s=2, t=2], they are assigned 0.33333... weight). And I got the result, which shows that BPT and reference will converge to the same. The minimum subpath lengths are set 1 there. com...
Tue Aug 21, 2012 1:36 pm
Forum: General Development
Topic: MIS BPT with Russian Roulette
Replies: 16
Views: 13728

### Re: MIS BPT with Russian Roulette

My first post contained misleading sentence. I attached three images that only represent 2 bounce transport, the first is the reference by PT, the second is the BPT which doesn't use RR, the third is the BPT which uses RR (strictly, minimum subpath lengths are set 3). The first twos look the same, b...
Tue Aug 21, 2012 12:04 pm
Forum: General Development
Topic: MIS BPT with Russian Roulette
Replies: 16
Views: 13728

### Re: MIS BPT with Russian Roulette

Thank you spectral. As you say, I tested getting rid of very small G factor by your code. As we expected, fireflies seem to be decreased, but whole of the image of BPT using RR seems to be brighter as before. Now I'm testing with minimum absolute path length (it is 1) for convenience. In practice, o...
Tue Aug 21, 2012 3:14 am
Forum: General Development
Topic: MIS BPT with Russian Roulette
Replies: 16
Views: 13728

### Re: MIS BPT with Russian Roulette

Thank you for your prompt reply. As you say, now I'm testing RR with a constant probability 0.5. The first few bounces don't consider RR, absolutely continue (but in the above code, RR is considered from the first bounce). I took RR into account in MIS weight by the below code. The precomputed PDFs ...
Sun Aug 19, 2012 5:18 pm
Forum: General Development
Topic: MIS BPT with Russian Roulette
Replies: 16
Views: 13728

### MIS BPT with Russian Roulette

Hello. Now, I'm implementing Veach-style BPT. And I could get a result which converge to the reference image generated by using path tracing. But this is only in case that I don't use Russian roulette in BPT. For debugging BPT, of course I checked the equality between PT and BPT's unweighted contrib...