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
Impossible to run #10
Comments
I solve this with replacement
to
But still hard to run. |
I solve above from [https://github.com//issues/3] a errors occur when i run scanNetCross.py some pixels of label bigger than 256 (about 300+-) |
I managed to run it. but the performance is unreproducible I validate it on 107 scenes.
and result
what's wrong? |
Hi, Please check our test log here. The test log is tested with the provided pretrained model on 4 GPUs, if you use one GPU please modify the test_repeat number to be 4 times as original one. |
Thanks for reply. |
The repeated process is borrowed from SparseConvNet. The reason for adopting this strategy is that the input data to the 3D sub-network is voxelized while we need per-point predictions for the benchmark submission. Directly propagating voxel labels to points may lead to some errors, so we repeat the process to take more points into account during the voxelization process. |
I could find rotation, translation augmentation from 'dataset/voxelizer.py' and the arguments transmitted from 'dataset/scanNet3D.py'(parent of 'dataset/scanNetCross,py') [rotation augmentation + translation augmentation + soft voting] That was why this model have coarser resolution than 5cm. Now i get it. Thanks! But not i get another curiosity about performance comparison. I'm not talking you have done unfair comparison. |
This repeat testing trick is borrowed from SparseConvNet, the SpaseConv-based competitors also adopt this trick, e.g., our baseline (MinkowskiUNet). So we think the comparison is fair. |
I do not want to train, but only want to test.
I download pretrained "bpnet_5cm.pth.tar"
and Set my config
And
$ sh tool/test.sh EXP1 ./config/scannet/bpnet_5cm.yaml 8
And get
I've already solved too many problems to get this error.
I sure even if i solve this error, another error will occur.
My conclusion is that you have not tested this on other environment at all.
Please test your opensource on other environment.
And give more details on README.md about how to run this.
Thank you.
The text was updated successfully, but these errors were encountered: