Rename ValidConvolutions to SubmanifoldConvolutions, update for PyTorch 0.4 Tensor/Variable merge

Torch/JoinTable.lua

@@ -7,7 +7,7 @@
 --[[
 Assume all the inputs have identical SparseGrids and input[i].features:size(1)
 Assume input[1].features:size(2) >= input[i].features:size(2) for all i=1,#input
-output.validRules is taken from input[1].validRules (could do set union?)
+output.submanifoldRules is taken from input[1].submanifoldRules (could do set union?)
 (for resnets, make sure the residual link is input[2])
 ]]
 

Torch/Metadata.lua

@@ -8,7 +8,7 @@
 Store Metadata relating to which spatial locations are active at each scale.
 Convolutions and 'convolution reversing' deconvolutions
 all coexist within the same MetaData object as long as each spatial size
-only occurs once. Valid convolutions do not change the spatial structure.
+only occurs once. Submanifold convolutions do not change the spatial structure.
 
 Serialization is emulated by storing the pointer as an integer.
 This is sufficient for mutithreaded batch preparation: each 'serialized'

Torch/NetworkArchitectures.lua

@@ -45,7 +45,7 @@ return function(sparseconvnet)
   function sparseconvnet.SparseVggNet(dimension,nInputPlanes,layers,opts)
     --[[
     VGG style nets
-    Use valid convolutions
+    Use submanifold convolutions
     Also implements 'Plus'-augmented nets
     ]]
     local nPlanes=nInputPlanes
@@ -57,20 +57,20 @@ return function(sparseconvnet)
       elseif x[1] == 'MP' then
         m:add(sparseconvnet.MaxPooling(dimension,x[2],x[3]))
       elseif x[1]=='C' and #x==2 then
-        m:add(sparseconvnet.ValidConvolution(dimension,nPlanes,x[2],3,false))
+        m:add(sparseconvnet.SubmanifoldConvolution(dimension,nPlanes,x[2],3,false))
         nPlanes=x[2]
         m:add(sparseconvnet.BatchNormReLU(nPlanes))
       elseif x[1]=='C' and #x==3 then
         m:add(sparseconvnet.ConcatTable()
           :add(
             sparseconvnet.Sequential()
-            :add(sparseconvnet.ValidConvolution(dimension,nPlanes,x[2],3,false))
+            :add(sparseconvnet.SubmanifoldConvolution(dimension,nPlanes,x[2],3,false))
           )
           :add(
             sparseconvnet.Sequential()
             :add(sparseconvnet.Convolution(dimension,nPlanes,x[3],3,2,false))
             :add(sparseconvnet.BatchNormReLU(x[3]))
-            :add(sparseconvnet.ValidConvolution(dimension,x[3],x[3],3,false))
+            :add(sparseconvnet.SubmanifoldConvolution(dimension,x[3],x[3],3,false))
             :add(sparseconvnet.BatchNormReLU(x[3]))
             :add(sparseconvnet.Deconvolution(dimension,x[3],x[3],3,2,false))
         ))
@@ -81,28 +81,28 @@ return function(sparseconvnet)
         m:add(sparseconvnet.ConcatTable()
           :add(
             sparseconvnet.Sequential()
-            :add(sparseconvnet.ValidConvolution(dimension,nPlanes,x[2],3,false))
+            :add(sparseconvnet.SubmanifoldConvolution(dimension,nPlanes,x[2],3,false))
           )
           :add(
             sparseconvnet.Sequential()
             :add(sparseconvnet.Convolution(dimension,nPlanes,x[3],3,2,false))
             :add(sparseconvnet.BatchNormReLU(x[3]))
-            :add(sparseconvnet.ValidConvolution(dimension,x[3],x[3],3,false))
+            :add(sparseconvnet.SubmanifoldConvolution(dimension,x[3],x[3],3,false))
             :add(sparseconvnet.BatchNormReLU(x[3]))
             :add(sparseconvnet.Deconvolution(dimension,x[3],x[3],3,2,false))
           )
           :add(sparseconvnet.Sequential()
             :add(sparseconvnet.Convolution(dimension,nPlanes,x[4],3,2,false))
             :add(sparseconvnet.BatchNormReLU(x[4]))
-            :add(sparseconvnet.ValidConvolution(dimension,x[4],x[4],3,false))
+            :add(sparseconvnet.SubmanifoldConvolution(dimension,x[4],x[4],3,false))
             :add(sparseconvnet.BatchNormReLU(x[4]))
             :add(sparseconvnet.Convolution(dimension,x[4],x[4],3,2,false))
             :add(sparseconvnet.BatchNormReLU(x[4]))
-            :add(sparseconvnet.ValidConvolution(dimension,x[4],x[4],3,false))
+            :add(sparseconvnet.SubmanifoldConvolution(dimension,x[4],x[4],3,false))
             :add(sparseconvnet.BatchNormReLU(x[4]))
             :add(sparseconvnet.Deconvolution(dimension,x[4],x[4],3,2,false))
             :add(sparseconvnet.BatchNormReLU(x[4]))
-            :add(sparseconvnet.ValidConvolution(dimension,x[4],x[4],3,false))
+            :add(sparseconvnet.SubmanifoldConvolution(dimension,x[4],x[4],3,false))
             :add(sparseconvnet.BatchNormReLU(x[4]))
             :add(sparseconvnet.Deconvolution(dimension,x[4],x[4],3,2,false))
         ))
@@ -138,10 +138,10 @@ return function(sparseconvnet)
               :add( --convolutional connection
                 sparseconvnet.Sequential()
                 :add(stride==1 and
-                  sparseconvnet.ValidConvolution(dimension,nPlanes,n,3,false) or
+                  sparseconvnet.SubmanifoldConvolution(dimension,nPlanes,n,3,false) or
                   sparseconvnet.Convolution(dimension,nPlanes,n,3,stride,false))
                 :add(sparseconvnet.BatchNormReLU(n))
-                :add(sparseconvnet.ValidConvolution(
+                :add(sparseconvnet.SubmanifoldConvolution(
                     dimension,n,n,3,false)))
               :add(residual(nPlanes,n,stride))
             )
@@ -150,10 +150,10 @@ return function(sparseconvnet)
               :add( --convolutional connection
                 sparseconvnet.Sequential()
                 :add(sparseconvnet.BatchNormReLU(n))
-                :add(sparseconvnet.ValidConvolution(
+                :add(sparseconvnet.SubmanifoldConvolution(
                     dimension,nPlanes,n,3,false))
                 :add(sparseconvnet.BatchNormReLU(n))
-                :add(sparseconvnet.ValidConvolution(
+                :add(sparseconvnet.SubmanifoldConvolution(
                     dimension,n,n,3,false))
               )
               :add(sparseconvnet.Identity())
@@ -169,7 +169,7 @@ return function(sparseconvnet)
   end
   function sparseconvnet.SparseDenseNet(dimension,nInputPlanes,layers)
     --[[
-    SparseConvNet meets DenseNets using valid convolutions
+    SparseConvNet meets DenseNets using submanifold convolutions
     Could do with a less confusing name
     ]]
     local nPlanes=nInputPlanes

Torch/ValidConvolution.lua → Torch/SubmanifoldConvolution.lua

@@ -8,7 +8,7 @@ return function(sparseconvnet)
   local C = sparseconvnet.C
 
   local Convolution, parent = torch.class(
-    'sparseconvnet.ValidConvolution', 'nn.Module', sparseconvnet)
+    'sparseconvnet.SubmanifoldConvolution', 'nn.Module', sparseconvnet)
 
   function Convolution:__init(dimension, nInputPlanes, nOutputPlanes,
       filterSize, bias)
@@ -52,7 +52,7 @@ return function(sparseconvnet)
     self.output.spatialSize = input.spatialSize
     self.shared.forwardPassMultiplyAddCount=
     self.shared.forwardPassMultiplyAddCount+
-    C.dimTypedFn(self.dimension, self._type, 'ValidConvolution_updateOutput')(
+    C.dimTypedFn(self.dimension, self._type, 'SubmanifoldConvolution_updateOutput')(
       input.spatialSize:cdata(),
       self.filterSize:cdata(),
       input.metadata.ffi,
@@ -68,7 +68,7 @@ return function(sparseconvnet)
   end
 
   function Convolution:backward(input, gradOutput)
-    C.dimTypedFn(self.dimension, self._type, 'ValidConvolution_backward')(
+    C.dimTypedFn(self.dimension, self._type, 'SubmanifoldConvolution_backward')(
       input.spatialSize:cdata(),
       self.filterSize:cdata(),
       input.metadata.ffi,
@@ -100,7 +100,7 @@ return function(sparseconvnet)
   end
 
   function Convolution:__tostring()
-    local s = 'ValidConvolution ' .. self.nInputPlanes .. '->' .. self.nOutputPlanes..' C'
+    local s = 'SubmanifoldConvolution ' .. self.nInputPlanes .. '->' .. self.nOutputPlanes..' C'
     if self.filterSize:max()==self.filterSize:min() and
     self.filterStride:max()==self.filterStride:min() then
       s=s..self.filterSize[1] ..(self.filterStride[1]==1 and

Torch/init.lua

@@ -35,7 +35,7 @@ for _,module in ipairs({
     'sparseconvnet.ReLU',
     'sparseconvnet.Sequential',
     'sparseconvnet.SparseToDense',
-    'sparseconvnet.ValidConvolution',
+    'sparseconvnet.SubmanifoldConvolution',
   }) do
   require(module)(sparseconvnet)
 end

examples/Assamese_handwriting/DenseNet.lua

@@ -14,7 +14,7 @@ local sparseModel = sparseconvnet.Sequential()
 local denseModel = nn.Sequential()
 model:add(sparseModel):add(denseModel)
 sparseModel
-:add(sparseconvnet.ValidConvolution(2,3,16,3,false))
+:add(sparseconvnet.SubmanifoldConvolution(2,3,16,3,false))
 :add(sparseconvnet.SparseDenseNet(2,16,{
       {'MP',compression=0},
       {nExtraLayers=2, growthRate=16},

examples/Assamese_handwriting/DenseNet.py

@@ -17,7 +17,7 @@ model = nn.Sequential()
 sparseModel = scn.Sequential()
 denseModel = nn.Sequential()
 model.add(sparseModel).add(denseModel)
-sparseModel.add(scn.ValidConvolution(2, 3, 8, 3, False))\
+sparseModel.add(scn.SubmanifoldConvolution(2, 3, 8, 3, False))\
     .add(scn.SparseDenseNet(2, 8, [
         {'pool': 'MP', 'compression': 0},
         {'nExtraLayers': 2, 'growthRate': 8},

examples/Assamese_handwriting/ResNet.lua

@@ -14,7 +14,7 @@ local sparseModel = sparseconvnet.Sequential()
 local denseModel = nn.Sequential()
 model:add(sparseModel):add(denseModel)
 sparseModel
-:add(sparseconvnet.ValidConvolution(2,3,8,3,false))
+:add(sparseconvnet.SubmanifoldConvolution(2,3,8,3,false))
 :add(sparseconvnet.MaxPooling(2,3,2))
 :add(sparseconvnet.SparseResNet(
     2,8,{

examples/Assamese_handwriting/ResNet.py

@@ -10,11 +10,13 @@ import sparseconvnet as scn
 from data import getIterators
 
 # two-dimensional SparseConvNet
+
+
 class Model(nn.Module):
     def __init__(self):
         nn.Module.__init__(self)
-        self.sparseModel=scn.Sequential(
-        ).add(scn.ValidConvolution(2, 3, 8, 3, False)
+        self.sparseModel = scn.Sequential(
+        ).add(scn.SubmanifoldConvolution(2, 3, 8, 3, False)
               ).add(scn.MaxPooling(2, 3, 2)
                     ).add(scn.SparseResNet(2, 8, [
                         ['b', 8, 2, 1],
@@ -23,15 +25,17 @@ class Model(nn.Module):
                         ['b', 32, 2, 2]])
         ).add(scn.Convolution(2, 32, 64, 5, 1, False)
               ).add(scn.BatchNormReLU(64)
-        ).add(scn.SparseToDense(2,64))
+                    ).add(scn.SparseToDense(2, 64))
         self.linear = nn.Linear(64, 183)
+
     def forward(self, x):
         x = self.sparseModel(x)
-        x = x.view(-1,64)
+        x = x.view(-1, 64)
         x = self.linear(x)
         return x
 
-model=Model()
+
+model = Model()
 spatial_size = model.sparseModel.input_spatial_size(torch.LongTensor([1, 1]))
 print('Input spatial size:', spatial_size)
 dataset = getIterators(spatial_size, 63, 3)
@@ -42,4 +46,4 @@ scn.ClassificationTrainValidate(
      'lr_decay': 0.05,
      'weight_decay': 1e-4,
      'use_gpu': torch.cuda.is_available(),
-    'check_point': True,})
+     'check_point': True, })

examples/Assamese_handwriting/ResNet_legacy.py

@@ -17,7 +17,7 @@ model = nn.Sequential()
 sparseModel = scn.Sequential()
 denseModel = nn.Sequential()
 model.add(sparseModel).add(denseModel)
-sparseModel.add(scn.ValidConvolution(2, 3, 8, 3, False))
+sparseModel.add(scn.SubmanifoldConvolution(2, 3, 8, 3, False))
 sparseModel.add(scn.MaxPooling(2, 3, 2))
 sparseModel.add(scn.SparseResNet(2, 8, [
     ['b', 8, 2, 1],

examples/Assamese_handwriting/VGGplus.py

@@ -16,19 +16,21 @@ class Model(nn.Module):
         self.sparseModel = scn.SparseVggNet(2, 3, [
             ['C', 8, ], ['C', 8], 'MP',
             ['C', 16], ['C', 16], 'MP',
-            ['C', 16, 8], ['C', 16, 8], 'MP',
-            ['C', 24, 8], ['C', 24, 8], 'MP']
+            ['C', 16 + 8], ['C', 16 + 8], 'MP',
+            ['C', 24 + 8], ['C', 24 + 8], 'MP']
         ).add(scn.Convolution(2, 32, 64, 5, 1, False)
               ).add(scn.BatchNormReLU(64)
-            ).add(scn.SparseToDense(2,64))
+                    ).add(scn.SparseToDense(2, 64))
         self.linear = nn.Linear(64, 183)
+
     def forward(self, x):
         x = self.sparseModel(x)
-        x = x.view(-1,64)
+        x = x.view(-1, 64)
         x = self.linear(x)
         return x
 
-model=Model()
+
+model = Model()
 spatial_size = model.sparseModel.input_spatial_size(torch.LongTensor([1, 1]))
 print('Input spatial size:', spatial_size)
 dataset = getIterators(spatial_size, 63, 3)
@@ -39,4 +41,4 @@ scn.ClassificationTrainValidate(
      'lr_decay': 0.05,
      'weight_decay': 1e-4,
      'use_gpu': torch.cuda.is_available(),
-    'check_point': True,})
+     'check_point': True, })

examples/Assamese_handwriting/VGGplus_legacy.py

@@ -33,6 +33,10 @@ print(model)
 spatial_size = sparseModel.suggestInputSize(torch.LongTensor([1, 1]))
 print('input spatial size', spatial_size)
 dataset = getIterators(spatial_size, 63, 3)
-scn.ClassificationTrainValidate(
-    model, dataset,
-    {'nEpochs': 100, 'initial_LR': 0.1, 'LR_decay': 0.05, 'weightDecay': 1e-4, 'checkPoint': False})
+scn.ClassificationTrainValidate(model,
+                                dataset,
+                                {'nEpochs': 100,
+                                 'initial_LR': 0.1,
+                                 'LR_decay': 0.05,
+                                 'weightDecay': 1e-4,
+                                 'checkPoint': False})

examples/Assamese_handwriting/data.py

@@ -10,7 +10,7 @@ import sparseconvnet.legacy as scn
 import pickle
 import math
 import random
-import numpy
+import numpy as np
 import os
 
 if not os.path.exists('pickle/'):
@@ -25,6 +25,8 @@ def train(spatial_size, Scale, precomputeStride):
     for i in range(9):
         for j in range(6588):
             d.append(d[j])
+    for i, x in enumerate(d):
+        x['idx'] = i
     d = torchnet.dataset.ListDataset(d)
     randperm = torch.randperm(len(d))
 
@@ -36,10 +38,11 @@ def train(spatial_size, Scale, precomputeStride):
         center = spatial_size.float().view(1, 2) / 2
         p = torch.LongTensor(2)
         v = torch.FloatTensor([1, 0, 0])
+        np_random = np.random.RandomState(tbl['idx'])
         for char in tbl['input']:
             inp.addSample()
             m = torch.eye(2)
-            r = random.randint(1, 3)
+            r = np_random.randint(1, 3)
             alpha = random.uniform(-0.2, 0.2)
             if alpha == 1:
                 m[0][1] = alpha
@@ -74,7 +77,7 @@ def train(spatial_size, Scale, precomputeStride):
                 #     v[1]=(x2-x1)/l
                 #     v[2]=(y2-y1)/l
                 #     l=max(x2-x1,y2-y1,x1-x2,y1-y2,0.9)
-                #     for j in numpy.arange(0,1,1/l):
+                #     for j in np.arange(0,1,1/l):
                 #         p[0]=math.floor(x1*j+x2*(1-j))
                 #         p[1]=math.floor(y1*j+y2*(1-j))
                 #         inp.setLocation(p,v,False)

examples/Chinese_handwriting/DenseNet-A-VSC.lua

@@ -14,7 +14,7 @@ local denseModel = nn.Sequential()
 local model = nn.Sequential():add(sparseModel):add(denseModel)
 
 sparseModel
-:add(sparseconvnet.ValidConvolution(2,3,16,3,false))
+:add(sparseconvnet.SubmanifoldConvolution(2,3,16,3,false))
 :add(sparseconvnet.SparseDenseNet(2,16,{
       {'MP',compression=0},
       {nExtraLayers=2, growthRate=16},

examples/Chinese_handwriting/DenseNet-A-VSC.py

@@ -17,7 +17,7 @@ model = nn.Sequential()
 sparseModel = scn.Sequential()
 denseModel = nn.Sequential()
 model.add(sparseModel).add(denseModel)
-sparseModel.add(scn.ValidConvolution(2, 3, 16, 3, False))\
+sparseModel.add(scn.SubmanifoldConvolution(2, 3, 16, 3, False))\
     .add(scn.SparseDenseNet(2, 16, [
         {'pool': 'MP', 'compression': 0},
         {'nExtraLayers': 2, 'growthRate': 16},

examples/Chinese_handwriting/DenseNet-B-VSC.lua

@@ -14,7 +14,7 @@ local denseModel = nn.Sequential()
 local model = nn.Sequential():add(sparseModel):add(denseModel)
 
 sparseModel
-:add(sparseconvnet.ValidConvolution(2,3,16,3,false))
+:add(sparseconvnet.SubmanifoldConvolution(2,3,16,3,false))
 :add(sparseconvnet.SparseDenseNet(2,16,{
       {'MP',compression=0},
       {nExtraLayers=4, growthRate=16},

examples/Chinese_handwriting/ResNet-A-VSC.lua

@@ -24,7 +24,7 @@ local denseModel = nn.Sequential()
 local model = nn.Sequential():add(sparseModel):add(denseModel)
 
 sparseModel
-:add(sparseconvnet.ValidConvolution(2,3,16,3,false))
+:add(sparseconvnet.SubmanifoldConvolution(2,3,16,3,false))
 :add(sparseconvnet.MaxPooling(2,2,2))
 :add(sparseconvnet.SparseResNet(
     2,16,{

examples/Chinese_handwriting/ResNet-A-VSC.py

@@ -17,7 +17,7 @@ model = nn.Sequential()
 sparseModel = scn.Sequential()
 denseModel = nn.Sequential()
 model.add(sparseModel).add(denseModel)
-sparseModel.add(scn.ValidConvolution(2, 3, 16, 3, False))
+sparseModel.add(scn.SubmanifoldConvolution(2, 3, 16, 3, False))
 sparseModel.add(scn.MaxPooling(2, 3, 2))
 sparseModel.add(scn.SparseResNet(2, 16, [
     ['b', 16, 2, 1],

examples/Chinese_handwriting/ResNet-B-VSC.lua

@@ -14,7 +14,7 @@ local denseModel = nn.Sequential()
 local model = nn.Sequential():add(sparseModel):add(denseModel)
 
 sparseModel
-:add(sparseconvnet.ValidConvolution(2,3,16,3,false))
+:add(sparseconvnet.SubmanifoldConvolution(2,3,16,3,false))
 :add(sparseconvnet.MaxPooling(2,2,2))
 :add(sparseconvnet.SparseResNet(
     2,16,{

examples/Chinese_handwriting/ResNet-C-VSC.lua

@@ -24,7 +24,7 @@ local denseModel = nn.Sequential()
 local model = nn.Sequential():add(sparseModel):add(denseModel)
 
 sparseModel
-:add(sparseconvnet.ValidConvolution(2,3,16,3,false))
+:add(sparseconvnet.SubmanifoldConvolution(2,3,16,3,false))
 :add(sparseconvnet.MaxPooling(2,2,2))
 :add(sparseconvnet.SparseResNet(
     2,16,{