Thomas Germer

Results 141 comments of Thomas Germer

> In NumPy 1.20.0, they [deprecated a series of aliases](https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations). I believe that the argument for deprecating inbuilt types in NumPy is a bit stronger since it was know for...

Another example: Why does the network decide to pick the sieve and not the people behind it? ![Screenshot_2020-06-19 Attention-Guided Hierarchical Structure Aggregation for Image Matting - Qiao_Attention-Guided_Hierarc (1)](https://user-images.githubusercontent.com/18725165/85106891-47f85f00-b20d-11ea-9e43-fcfb805ddf07.png) My guess...

I contacted the authors by e-mail and received a link to the dataset and can confirm that the alpha mattes are of very high quality. The subjects are mostly people,...

> I am happy to leave it open if you wish to continue on the PR and address my comment on the other one Are you referring to [this comment](https://github.com/numpy/numpy/pull/21805#discussion_r902615078)?...

## 2) Writable views Writing to the diagonal of a matrix with `np.einsum` still works for me. Perhaps you wrote `diagonal = ...` instead of `diagonal[:] = ...`? The later...

I thought so, too. I read the docs now and apparently almost the same example can be found further down the page. Maybe it is possible to link from the...

Here is a test case for a 16 bit PNG image generated with GIMP. ```python import io import base64 from PIL import Image # Create a PIL.Image from a base64-encoded...

Compiling and running with address sanitizer does not raise any issues: ```bash gcc -g -fsanitize=address -fsanitize=undefined -pipe -Wall -O3 -fomit-frame-pointer -march=native -std=c99 -mfpmath=sse -msse3 -fopenmp fasta.gcc-2.c -o fasta.gcc-2.gcc_run && ./fasta.gcc-2.gcc_run...

This optimization will probably become irrelevant with the [new string dtype](https://numpy.org/neps/nep-0055-string_dtype.html), so I think this is not worth pursuing anymore.

There only is GPU support for foreground estimation, but not for alpha matting itself. For foreground estimation, simply install [PyOpenCL or CuPy](https://github.com/pymatting/pymatting/blob/master/requirements_gpu.txt) and import the relevant foreground estimation method https://github.com/pymatting/pymatting/blob/b5f03a2c373cfe3ca6ba0f69a56e113ca3c5807d/tests/test_foreground.py#L13...