Carlo de Franchis

The test added in PR #206 shows that removing RANSAC from the function `s2p.sift.matches_on_rpc_roi` changes the number of matches. I'll come up with a Python implementation to replace the ransac...

You can run `s2p` without downloading the input images by prefixing the url of your input images with `/vsicurl/`. Support for GDAL vsicurl was introduced in pull request [#145](https://github.com/MISS3D/s2p/pull/145)

@jmichel-otb Right now it seems that running a specific test or group of tests doesn't work: python -m unittest tests.sift_test raises `ImportError: Failed to import test module: sift_test`

Actually I find the `unittest` framework heavy and hard to read because of all the boilerplate code. Would you mind if we use [pytest](https://docs.pytest.org/en/latest/index.html) instead?

Thank you @jmichel-otb for this detailed answer. I propose not to use any framework, neither `unittest` nor `pytest`. `pytest` can be used as a convenience tool to run the tests...

Thank you @amiotc. I've fixed a coordinates swap in the file `sift4ctypes.cpp` and changed the datatype of a `ctypes.POINTER` to `ndpointer` in `sift.py`. Could you check that everything's ok?

@amiotc Great, thank you! So now if @jmichel-otb and @dyoussef approve the pull request I think that we can merge.

Work in progress in pull request #189. We still need to implement image filtering before the call to `scipy.ndimage.affine_transform` to prevent aliasing.

Hi Krishna, If you can provide the commit hash of the version that worked fine it could help us track down the cause of the difference. Best

Hi Krishna, Thank you very much for raising this issue. After a bit of analysis, it turns out that the problem is caused by an optimisation that was introduced in...