Torch crashes (out of bounds) with 3D model when there is more than one Concat module involved
Timo
Hello,
I try to create a model for 3D object classification. The simplest version works (no rotation, no scale only one block) [1], with the input (16x16x16) and output dimensions(16x32x16) [2]. When I add a second block [3] it crashes with this strange error [4] (bad argument #2 to '?' (out of bounds...)).
I'm new to torch so it is hard for me to make sense out of this. Is this an error of the model?
Thanks!
----------------------------------------------------------------
[1]
nn.Sequential {
[input -> (1) -> (2) -> (3) -> (4) -> (5) -> (6) -> (7) -> (8) -> output]
(1): nn.Parallel {
input
`-> (1): nn.Sequential {
[input -> (1) -> (2) -> output]
(1): nn.Parallel {
input
`-> (1): nn.Sequential {
[input -> (1) -> output]
(1): nn.Concat {
input
|`-> (1): nn.Sequential {
| [input -> (1) -> output]
| (1): nn.VolumetricConvolution(1 -> 4, 3x3x3, 1,1,1, 1,1,1)
| }
`-> (2): nn.Sequential {
[input -> (1) -> output]
(1): nn.VolumetricConvolution(1 -> 4, 1x1x1)
}
... -> output
}
}
... -> output
}
(2): nn.Reshape(1x32768)
}
... -> output
}
(2): nn.SpatialMaxPooling(1x1, 1,1)
(3): nn.Reshape(32768)
(4): nn.Linear(32768 -> 512)
(5): nn.ReLU
(6): nn.Dropout(0.500000)
(7): nn.Linear(512 -> 8)
(8): nn.LogSoftMax
}
----------------------------------------------------------------
[2]
nn.Parallel {
input
`-> (1): nn.Sequential {
[input -> (1) -> output]
(1): nn.Concat {
input
|`-> (1): nn.Sequential {
| [input -> (1) -> output]
| (1): nn.VolumetricConvolution(1 -> 4, 3x3x3, 1,1,1, 1,1,1)
| }
`-> (2): nn.Sequential {
[input -> (1) -> output]
(1): nn.VolumetricConvolution(1 -> 4, 1x1x1)
}
... -> output
}
}
... -> output
}
1000
4
16
32
16
[torch.LongStorage of size 5]
nn.Sequential {
[input -> (1) -> output]
(1): nn.Concat {
input
|`-> (1): nn.Sequential {
| [input -> (1) -> output]
| (1): nn.VolumetricConvolution(1 -> 4, 3x3x3, 1,1,1, 1,1,1)
| }
`-> (2): nn.Sequential {
[input -> (1) -> output]
(1): nn.VolumetricConvolution(1 -> 4, 1x1x1)
}
... -> output
}
}
1000
4
16
32
16
[torch.LongStorage of size 5]
----------------------------------------------------------------
[3]
nn.Sequential {
[input -> (1) -> (2) -> (3) -> (4) -> (5) -> (6) -> (7) -> (8) -> output]
(1): nn.Parallel {
input
`-> (1): nn.Sequential {
[input -> (1) -> (2) -> output]
(1): nn.Parallel {
input
`-> (1): nn.Sequential {
[input -> (1) -> (2) -> output]
(1): nn.Concat {
input
|`-> (1): nn.Sequential {
| [input -> (1) -> output]
| (1): nn.VolumetricConvolution(1 -> 4, 3x3x3, 1,1,1, 1,1,1)
| }
`-> (2): nn.Sequential {
[input -> (1) -> output]
(1): nn.VolumetricConvolution(1 -> 4, 1x1x1)
}
... -> output
}
(2): nn.Concat {
input
|`-> (1): nn.Sequential {
| [input -> (1) -> output]
| (1): nn.VolumetricConvolution(4 -> 8, 3x3x3, 1,1,1, 1,1,1)
| }
`-> (2): nn.Sequential {
[input -> (1) -> output]
(1): nn.VolumetricConvolution(4 -> 8, 1x1x1)
}
... -> output
}
}
... -> output
}
(2): nn.Reshape(1x131072)
}
... -> output
}
(2): nn.SpatialMaxPooling(1x1, 1,1)
(3): nn.Reshape(131072)
(4): nn.Linear(131072 -> 1024)
(5): nn.ReLU
(6): nn.Dropout(0.500000)
(7): nn.Linear(1024 -> 8)
(8): nn.LogSoftMax
}
----------------------------------------------------------------
[4]
bad argument #2 to '?' (out of bounds at /home/timo/src/torch/pkg/torch/lib/TH/generic/THStorage.c:202)
stack traceback:
[C]: at 0x7f4000c30420
[C]: in function '__index'
/home/timo/src/torch/install/share/lua/5.1/nn/Concat.lua:19: in function </home/timo/src/torch/install/share/lua/5.1/nn/Concat.lua:9>
[C]: in function 'xpcall'
...timo/src/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors'
...timo/src/torch/install/share/lua/5.1/nn/Sequential.lua:44: in function <...timo/src/torch/install/share/lua/5.1/nn/Sequential.lua:41>
[C]: in function 'xpcall'
...timo/src/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors'
.../timo/src/torch/install/share/lua/5.1/nn/Parallel.lua:18: in function <.../timo/src/torch/install/share/lua/5.1/nn/Parallel.lua:10>
[C]: in function 'xpcall'
...
[C]: in function 'xpcall'
...timo/src/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors'
.../timo/src/torch/install/share/lua/5.1/nn/Parallel.lua:18: in function <.../timo/src/torch/install/share/lua/5.1/nn/Parallel.lua:10>
[C]: in function 'xpcall'
...timo/src/torch/install/share/lua/5.1/nn/Container.lua:63: in function 'rethrowErrors'
...timo/src/torch/install/share/lua/5.1/nn/Sequential.lua:44: in function 'forward'
debug_model.lua:24: in main chunk
[C]: in function 'dofile'
.../src/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
[C]: at 0x00406670