cuda check

README.md

@@ -124,7 +124,9 @@ python VGGplus.py
 Tested with CUDA 10.0, Ubuntu 18.04, Python 3.6 with [Conda](https://www.anaconda.com/) and PyTorch 1.1.
 
 ```
-conda install pytorch-nightly -c pytorch # See https://pytorch.org/get-started/locally/
+conda install pytorch torchvision cudatoolkit=10.0 -c pytorch # See https://pytorch.org/get-started/locally/
+conda install google-sparsehash -c bioconda
+conda install -c anaconda pillow
 git clone git@github.com:facebookresearch/SparseConvNet.git
 cd SparseConvNet/
 bash develop.sh
@@ -148,6 +150,7 @@ SparseConvNet is BSD licensed, as found in the LICENSE file.
 8. [Workshop on Learning to See from 3D Data, 2017](https://shapenet.cs.stanford.edu/iccv17workshop/) First place in the [semantic segmentation](https://shapenet.cs.stanford.edu/iccv17/) competition. [Report](https://arxiv.org/pdf/1710.06104)
 9. [3D Semantic Segmentation with Submanifold Sparse Convolutional Networks, 2017](https://arxiv.org/abs/1711.10275) Semantic segmentation for the ShapeNet Core55 and NYU-DepthV2 datasets, CVPR 2018
 10. [ScanNet 3D semantic label benchmark 2018](http://kaldir.vc.in.tum.de/scannet_benchmark/semantic_label_3d) 0.726 average IOU.
+11. [https://github.com/StanfordVL/MinkowskiEngine] MinkowskiEngine is an alternative implementation of SparseConvNet.
 
 ## Citations
 

develop.sh

@@ -6,8 +6,5 @@
 # LICENSE file in the root directory of this source tree.
 
 rm -rf build/ dist/ sparseconvnet.egg-info sparseconvnet_SCN*.so
-conda install google-sparsehash -c bioconda
-conda install -c anaconda pillow
-conda install scipy
 python setup.py develop
 python examples/hello-world.py

examples/ScanNet/data.py

@@ -8,6 +8,7 @@
 scale=20  #Voxel size = 1/scale
 val_reps=1 # Number of test views, 1 or more
 batch_size=32
+elastic_deformation=False
 
 import torch, numpy as np, glob, math, torch.utils.data, scipy.ndimage, multiprocessing as mp
 
@@ -19,11 +20,11 @@ VALID_CLASS_IDS = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 14, 16, 24, 2
 
 train,val=[],[]
 for x in torch.utils.data.DataLoader(
-      glob.glob('train/*.pth'),
+        glob.glob('train/*.pth')[::10],
         collate_fn=lambda x: torch.load(x[0]), num_workers=mp.cpu_count()):
     train.append(x)
 for x in torch.utils.data.DataLoader(
-      glob.glob('val/*.pth'),
+        glob.glob('val/*.pth')[::10],
         collate_fn=lambda x: torch.load(x[0]), num_workers=mp.cpu_count()):
     val.append(x)
 print('Training examples:', len(train))
@@ -60,6 +61,7 @@ def trainMerge(tbl):
         theta=np.random.rand()*2*math.pi
         m=np.matmul(m,[[math.cos(theta),math.sin(theta),0],[-math.sin(theta),math.cos(theta),0],[0,0,1]])
         a=np.matmul(a,m)
+        if elastic_deformation:
             a=elastic(a,6*scale//50,40*scale/50)
             a=elastic(a,20*scale//50,160*scale/50)
         m=a.min(0)

setup.py

@@ -8,6 +8,9 @@ import torch, os
 from torch.utils.cpp_extension import BuildExtension, CppExtension, CUDAExtension
 from setuptools import setup, find_packages
 
+if torch.cuda.is_available():
+    assert torch.matmul(torch.ones(2097153,2).cuda(),torch.ones(2,2).cuda()).min().item()==2, 'Please upgrade from CUDA 9.0 to CUDA 10.0+'
+
 this_dir = os.path.dirname(os.path.realpath(__file__))
 torch_dir = os.path.dirname(torch.__file__)
 conda_include_dir = '/'.join(torch_dir.split('/')[:-4]) + '/include'