ilyatrofimov/mtopdiv

[NeurIPS 2021] Manifold Topology Divergence: a Framework for Comparing Data Manifolds

mtopdiv's Issues

Colab

Hi
When running example.ipynb on colab I get this error /content/ripser-plusplus/ripserplusplus/libphmap.so: cannot open shared object file: No such file or directory

Here is what I do to in my notebook to install the dependencies and run the example

!git clone https://github.com/simonzhang00/ripser-plusplus.git
%cd ripser-plusplus
!python setup.py install

(In here not sure if I have to go back to original folder %cd .. but it gives me the same error either way)

!git clone https://github.com/IlyaTrofimov/MTopDiv.git
%cd MTopDiv
!python setup.py install

import numpy as np
import mtd

np.random.seed(7)
P = np.random.rand(1000, 2)
Q = np.random.rand(1000, 2)

barc = mtd.calc_cross_barcodes(P, Q, batch_size1 = 100, batch_size2 = 1000)

and this is where the error occurs

Thanks for the help !

Evaluation in terms of precision and recall seperately

Hi! Thank you for providing your amazing evaluation metric sourcecode!

I just wonder if I can evaluate the datas using MTopDiv in terms of precision and recall seperately and how to implement that.

Thank you.

Batch size 1 and 2

Hi
Is there a minimal amount of generated/true images point clouds to get this work correctly ?
In your experiment you are using 10^3 and 10^4 for ffhq, I was wondering if it possible to lower those numbers (like Bq=100 and Bp=1000 where p is the data distribution and q the model distribution )
Thanks