search

RobinMagnet / pyFM

Python implementation of Functional Maps
MIT License
95 stars 20 forks source link

Matching parameters tuning #2

Closed ihmc3jn09hk closed 3 years ago

ihmc3jn09hk commented 3 years ago

Hi @RobinMagnet

I tried to use custom data.zip and perform the FM matching. The match does not provide proper correspondence. Many matches are flipped in some direction, e.g. x-axis. I have attached the p2p result in the provided zip file. The attachment also includes the tested off files with the landmarks. In order to get correct/good result, what parameters should be used ??

I tried:

k1,k2 = 120,120 # Number of eigenvalues on source and Target
landmarks = np.loadtxt('data/mergedLandmarks.txt',dtype=int)[:15] # loading 5 landmarks

P.S. In order to read one of the off (contains vertices color), the file file_utils.py line 24 has to be changed as following which discard reading the vertex-color.

faces = [[int(x) for x in f.readline().strip().split()][1:4] for _ in range(n_faces)]
ihmc3jn09hk commented 3 years ago

I was trying to use purely the descriptors themselves to do the fitting which I did not use the landmarks. After a fix. It is working.