Andrea Tagliasacchi

Potential baselines associated with this dataset (might branch out to separate task): "HMR" → Kanazawa et al. "End-to-end recovery of human shape and pose" (removed some comments below, as the...

If the error message would have told me that, I would have saved 1h of my time :) Just left a trace for others that might run into this. @Conchylicultor...

> Tell me, please, whether such experiments were carried out? Does the network actually restore the shape exactly? Yes, the teardrop image can be restored in overfitting mode.

Yes, this is just an illustration image (we didn't have a dataset of fonts at hands). It's a one off and we fiddled with initialization until overfitting the model to...

yes, the github actions workflow is not the standard continuous integration mechanism. However, *I would kindly ask you to delete the comment above* (and I will do the same with...

Can you clarify your second point? (the bug)

ah yes, of course! Missed this one as testing with multiple point sizes was not possible. Now with the latest edits to the testing pipeline this would have been caught.

Sure On Sat, May 9, 2020, 8:56 PM Dominic Jack wrote: > If I'm renaming layers anyway, any chance I can address #314 > e.g. > PointNetConv2Layer -> PointnetConv2D >...

I'll have time to look at things tomorrow, finishing reviews today. Note helpers should NOT contain any project logic. It's basically project agnostic functionality that is useful whenever you have...

So load("modelnet/pointnet", ...) would give you the variant with "exactly the same data preprocessing as pointnet". But perhaps, for now, a separate file is sufficient, something like modelnet_dataset.py or modelnet.py....