leanderweber

Just noting this, so we remember: PR #88 contains some indexing functions which are tailored to the current non-batching implementation of metrics in the main branch. These functions would also...

Specifically for ResNet50, there is also a pretty nasty artifact in modified backprop explanations caused by the 1x1conv stride 2 downsampling shortcuts. Short of excluding those shortcuts, there is not...

The imports should be fixed. Was an issue with relative imports not having been replaced by absolute imports in base_batched.py Also renamed normalized_axes to normalise_axes in merged content. Not sure...

Hi @davor10105, thank you for working on the batched metrics! Regarding your questions above: 1. **Filtering of overlapping patches**. The reason for the current implementation is to avoid fitting a...

Hi @davor10105 @annahedstroem, sorry for the late reply! I think this issue comes down to interpretation of the metric, and goals for implementing it. A grid is efficient and easy...