Open jaypatravali opened 7 years ago
I came from the blog post as well and was also getting very different results too. I was using 640x480 left and right images straight from the New Tsukuba dataset. To get the same results as on Sourish's blog I needed to increase ndisp
to 100 (i.e. computeDisparityMapORB(100)
).
Most of the disparities I found were in the 40-70 pixels range. Using ndisp=20
as on the blog or ndisp=40
from the current code on GitHub, I was getting very noisy results where you couldn't really see the shapes from the images.
Hey,
Arrived from your blog, I dont see good/same results from a fresh clone of your repo.
I used the same tsubuka_l/r.png images from your blog to test it. Any specific instructions?