Sean I Young, PhD

Make Dice a ModuleMetric similar to IoU

5

## 🚀 Feature Make Dice a ModuleMetric similar to IoU ### Motivation Currently, the usability of Dice (functional metric) is no on par with that of (modular metric) IoU. ###...

Distribute via conda as well?

1

Distributing surfa via conda would be cool! — I have surfa as a prerequisite for [this](https://www.github.com/seannz/svr) project and surfa is the only package that is not conda-installable...

Hello, I’m trying to build the PSP core with 2x PSP rendering and texture. Would anyone have any pointers for me? In general, how would one configure the psp core...

Llama 2 perplexity results on wikitext2?

6

Hello, Would you happen to already have the OWQ (3–4 bits) perplexity results for Llama 2 models on the WikiText2 dataset? It'd be great have my results cross-checked against yours....

Currently, the model is trained on slice stacks with cleanly masked foreground regions. It would be useful to randomize /pollute the foreground mask to robustly reconstruct from imperfectly masked slice...

Currently, the ground truth volumes are flipped L–R only. However, it would be useful to have the volume flipped L–R, A–P, I–S for more robust handling of arbitrarily oriented volumes...

Currently, the slice stack is always cropped with a 8-voxel padding around the bounding box, which may not be robust. Crop size should be randomized during training.