Closed james77777778 closed 10 months ago
Patch coverage: 86.06%
and project coverage change: +0.01%
:tada:
Comparison is base (
b4019bc
) 83.63% compared to head (722a9d1
) 83.64%.
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
Related to keras-team/keras#18442
This PR has implemented
ops.map_coordinates
for all backends based on the PR from @mihirparadkar keras-team/keras-core#784It is challenge to obtain a jittable
map_coordinates
for tensorflow, but I managed to figure out the solution. The key is to usetf.unstack
to separate coordinates and form a list of tensor for subsequent operations.The unit test is borrowed from jax and has been simpified https://github.com/google/jax/blob/bcc545a69232e983ae31b0395f4972979f2789c0/tests/scipy_ndimage_test.py#L79
The standalone script:
Results: