Closed MaxMayya closed 4 years ago
Hello, Our ACAP implementation is not available now. If we prepare to release the code, I will notify you. By the way, you can check the original author(http://geometrylearning.com/lin/) website to see whether they share the code.
Update
ACAP authors have finally published there linux implementation here.
I want to test that on the dataset you share; for ex. I want to reproduce dataset/neutral_fs/SPRING2647.txt
from dataset/neutral_ms/SPRING2647.obj
and the reference mesh. Could you please point me to the reference mesh used to generate the ACAP features please?
Actually, the reference mesh is not in the dataset. The reference mesh 'template.obj' can be found in the downloaded trained model(See Install). For some reason, we choose this non-neutral shape mesh as reference mesh, this is not very suitable. You can choose another neutral mesh as reference mesh and recompute your acap feature. And another note, different mixed integer optimization setting can lead to different acap values, even with same ref mesh. So, you may get different acap feature with their code.
Yes different optimizers could end up with different results I already noticed such behaviour on their test vectors. Out of curiosity, what optimizer do you use? Gurobi?
Yes. We also use gurobi.
Thank you for sharing this very interesting labor. I want to test your model on a specific dataset but I cannot find the implementation of ACAP anywhere. Could you please point me to some code?
Many thanks