Allen Goodman

@mmuckley This is surprisingly tricky! I could not find a Python package that does this flexibly. Internally this is performed by the FBSource Linters but that is, unfortunately, not open...

@mmuckley I went ahead and did the initial fixes for Python sources. See #148.

Hi, @system123. I have not generated ImageNet weights for ResNet-18. If you don't mind waiting a day or two, I can start creating them today. Let me know!

Thanks, @mjdietzx. Huge help!

I started work on napari/napari#269. Labels should, in my opinion, use the representation that is both ubiquitous in computer vision research and machine learning libraries like PyTorch and TensorFlow: `(n,...

@lassoan For sure. This _was_ the common structure in computer vision too. But this changed, like everything else in the past decade, when learned-based methods became standard. Think about overlapping...

@lassoan Your comment is really interesting! I should confess that I know absolutely nothing about microscopy! > I don't remember anyone asking for a 6th dimension in the past 10...

@constantinpape I have not followed this (or any other ngff) discussion until yesterday! I apologize for missing some important context. 😄 > I agree that being able to represent probabilistic...

> Trends can change quickly though, so I agree that the file format should be able to handle fractional labelmaps well. @lassoan Exactly. `argmax` predictions are, and I assume will...

@jhung0 You’e really close. Two comments: - Use `keras.backend.mean` rather than `tensorflow.reduce_mean` - We need to implement a custom average precision function because `tensorflow.metrics.sparse_average_precision_at_k` expects to be updated in training...