-
Notifications
You must be signed in to change notification settings - Fork 1.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Modelnet #194
Modelnet #194
Conversation
dataset info here. |
I have no idea why this presubmit is failing... can't install tensorflow-datasets? All I did was add a lazy import... |
Hi @jackd would you like to add this dataset under tensorflow_graphics instead? I finished the PR for modelnet40 (point cloud version, used by pointnet) and I randomly bumped in your pull request. |
@ataiya Yeah, I gave up waiting here/submitting any more PRs and just made a separate repository with this and some other shape stuff. If it's likely to get through at graphics then I'm happy to clean some of them up, but I wouldn't mind getting a clearer idea of what'll be prioritized at the other end before I make too many PRs like this one. |
... this was supposed to just be modelnet10/40 with tri-mesh and point cloud sampling... then I found this and extending it to their sampled data/shapenet parts didn't seem too much more. Apologies for not splitting it up.
One issue I haven't been able to resolve is repeated downloads for different configs of the same dataset.
builder.info.download_checksums
just seem to disappear occasionally between runs. Could be related to this, but I suspect it's a different issue.Note sure what the feeling is regarding
volume
submodule - just seemed like a good catch-all for point cloud/mesh/voxels. I suspect there'll be less controversial aspects here compared to this PR, but once either is resolved I'm happy to modify the other so that all shapenet stuff is sitting together (assuming image -> volume datasets are morevolume
-ous thanimage
-ish).Also happy to add the shapenet ICCV2017 reconstruction track (to partner the point cloud segmentation one) once shape issues with RLE encodings have been resolved.