Printing pretrained torch model options
engahmed1190 opened this issue · comments
I am trying to print the torch model options of training, I have used the provided model model.t7
I am using this code to print the option
require 'torch'
require 'nn'
require 'InstanceNormalization'
--[[
Prints the options that were used to train a a feedforward model.
--]]
local cmd = torch.CmdLine()
cmd:option('-model', 'models/instance_norm/candy.t7')
local opt = cmd:parse(arg)
print('Loading model from ' .. opt.model)
local checkpoint = torch.load(opt.model)
for k, v in pairs(checkpoint.opt) do
if type(v) == 'table' then
v = table.concat(v, ',')
end
print(string.format('%s: %s', k, v))
end
but I am getting this error:
/src/torch/install/bin/luajit: print_options.lua:22: bad argument #1 to 'pairs' (table expected, got nil)
stack traceback:
[C]: in function 'pairs'
print_options.lua:22: in main chunk
[C]: in function 'dofile'
.../src/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk
[C]: at 0x00406670
this is the model that you provided for reference , it loaded perfectly find but the main issue with the evaluate method .
nn.Sequential {
[input -> (0) -> (1) -> (2) -> (3) -> (4) -> (5) -> (6) -> (7) -> (8) -> (9) -> (10) -> (11) -> (12) -> (13) -> (14) -> (15) -> (16) -> (17) -> (18) -> (19) -> (20) -> (21) -> (22) -> (23) -> output]
(0): TorchObject(nn.TVLoss, {'_type': 'torch.FloatTensor', 'strength': 0, 'x_diff':
( 0 ,.,.) =
0.0000 0.0000 0.0000 ... -0.0039 0.0078 0.0000
0.0000 0.0000 0.0000 ... 0.0039 0.0000 0.0000
0.0000 0.0000 0.0000 ... 0.0078 -0.0157 0.0000
... ⋱ ...
-0.0196 0.0157 -0.0196 ... 0.0078 0.0588 0.0392
-0.0196 0.0196 -0.0235 ... 0.0235 0.0745 0.1451
-0.0078 0.0000 0.0118 ... 0.0431 0.1882 0.1569
( 1 ,.,.) =
0.0000 0.0000 0.0000 ... -0.0039 0.0078 0.0000
0.0000 0.0000 0.0000 ... 0.0039 0.0000 0.0000
0.0000 0.0000 0.0000 ... 0.0078 -0.0078 0.0000
... ⋱ ...
-0.0196 0.0157 -0.0196 ... 0.0078 0.0431 0.0235
-0.0196 0.0196 -0.0235 ... 0.0196 0.0588 0.1373
-0.0078 0.0000 0.0118 ... 0.0353 0.1765 0.1451
( 2 ,.,.) =
0.0000 0.0000 0.0000 ... -0.0039 0.0078 0.0000
0.0000 0.0000 0.0000 ... 0.0039 0.0000 0.0000
0.0000 0.0000 0.0000 ... 0.0078 -0.0078 0.0000
... ⋱ ...
-0.0196 0.0157 -0.0196 ... 0.0078 0.0471 0.0235
-0.0196 0.0196 -0.0235 ... 0.0118 0.0510 0.1216
-0.0078 0.0039 0.0118 ... 0.0157 0.1647 0.1176
[torch.FloatTensor of size 3x511x511]
, 'gradInput': [torch.FloatTensor with no dimension]
, 'y_diff':
( 0 ,.,.) =
0.0000 0.0000 0.0000 ... 0.0000 0.0078 0.0000
0.0000 0.0000 0.0000 ... 0.0078 0.0118 -0.0039
0.0000 0.0000 0.0000 ... -0.0118 -0.0078 0.0000
... ⋱ ...
0.0039 0.0039 0.0078 ... 0.0157 0.0314 0.0471
-0.0157 -0.0039 -0.0235 ... 0.0863 0.1059 0.2196
0.0039 0.0039 0.0039 ... 0.1020 0.1961 0.1451
( 1 ,.,.) =
0.0000 0.0000 0.0000 ... 0.0000 0.0078 0.0000
0.0000 0.0000 0.0000 ... 0.0000 0.0039 -0.0039
0.0000 0.0000 0.0000 ... -0.0118 -0.0078 0.0000
... ⋱ ...
0.0039 0.0039 0.0078 ... 0.0196 0.0314 0.0471
-0.0157 -0.0039 -0.0235 ... 0.0706 0.0863 0.2039
0.0039 0.0039 0.0039 ... 0.0941 0.1922 0.1451
( 2 ,.,.) =
0.0000 0.0000 0.0000 ... 0.0000 0.0078 0.0000
0.0000 0.0000 0.0000 ... 0.0000 0.0039 -0.0039
0.0000 0.0000 0.0000 ... -0.0196 -0.0078 0.0000
... ⋱ ...
0.0039 0.0039 0.0078 ... 0.0039 0.0078 0.0118
-0.0196 -0.0078 -0.0235 ... 0.0275 0.0314 0.1451
0.0039 0.0039 0.0039 ... 0.0745 0.1725 0.0980
[torch.FloatTensor of size 3x511x511]
, 'train': True, 'output': [torch.FloatTensor with no dimension]
})
(1): nn.SpatialReplicationPadding(4, 4, 4, 4)
(2): nn.SpatialConvolution(3 -> 32, 9x9)
(3): nn.InstanceNormalization
(4): nn.ReLU
(5): nn.SpatialConvolution(32 -> 64, 3x3, 2, 2, 1, 1)
(6): nn.InstanceNormalization
(7): nn.ReLU
(8): nn.SpatialConvolution(64 -> 128, 3x3, 2, 2, 1, 1)
(9): nn.InstanceNormalization
(10): nn.ReLU
(11): nn.Sequential {
[input -> (0) -> (1) -> output]
(0): torch.legacy.nn.ConcatTable.ConcatTable {
input
|`-> (0): nn.Identity
|`-> (1): nn.Sequential {
[input -> (0) -> (1) -> (2) -> (3) -> (4) -> (5) -> (6) -> output]
(0): nn.SpatialReplicationPadding(1, 1, 1, 1)
(1): nn.SpatialConvolution(128 -> 128, 3x3)
(2): nn.InstanceNormalization
(3): nn.ReLU
(4): nn.SpatialReplicationPadding(1, 1, 1, 1)
(5): nn.SpatialConvolution(128 -> 128, 3x3)
(6): nn.InstanceNormalization
}
+. -> output
}
(1): nn.CAddTable
}
(12): nn.Sequential {
[input -> (0) -> (1) -> output]
(0): torch.legacy.nn.ConcatTable.ConcatTable {
input
|`-> (0): nn.Identity
|`-> (1): nn.Sequential {
[input -> (0) -> (1) -> (2) -> (3) -> (4) -> (5) -> (6) -> output]
(0): nn.SpatialReplicationPadding(1, 1, 1, 1)
(1): nn.SpatialConvolution(128 -> 128, 3x3)
(2): nn.InstanceNormalization
(3): nn.ReLU
(4): nn.SpatialReplicationPadding(1, 1, 1, 1)
(5): nn.SpatialConvolution(128 -> 128, 3x3)
(6): nn.InstanceNormalization
}
+. -> output
}
(1): nn.CAddTable
}
(13): nn.Sequential {
[input -> (0) -> (1) -> output]
(0): torch.legacy.nn.ConcatTable.ConcatTable {
input
|`-> (0): nn.Identity
|`-> (1): nn.Sequential {
[input -> (0) -> (1) -> (2) -> (3) -> (4) -> (5) -> (6) -> output]
(0): nn.SpatialReplicationPadding(1, 1, 1, 1)
(1): nn.SpatialConvolution(128 -> 128, 3x3)
(2): nn.InstanceNormalization
(3): nn.ReLU
(4): nn.SpatialReplicationPadding(1, 1, 1, 1)
(5): nn.SpatialConvolution(128 -> 128, 3x3)
(6): nn.InstanceNormalization
}
+. -> output
}
(1): nn.CAddTable
}
(14): nn.Sequential {
[input -> (0) -> (1) -> output]
(0): torch.legacy.nn.ConcatTable.ConcatTable {
input
|`-> (0): nn.Identity
|`-> (1): nn.Sequential {
[input -> (0) -> (1) -> (2) -> (3) -> (4) -> (5) -> (6) -> output]
(0): nn.SpatialReplicationPadding(1, 1, 1, 1)
(1): nn.SpatialConvolution(128 -> 128, 3x3)
(2): nn.InstanceNormalization
(3): nn.ReLU
(4): nn.SpatialReplicationPadding(1, 1, 1, 1)
(5): nn.SpatialConvolution(128 -> 128, 3x3)
(6): nn.InstanceNormalization
}
+. -> output
}
(1): nn.CAddTable
}
(15): nn.Sequential {
[input -> (0) -> (1) -> output]
(0): torch.legacy.nn.ConcatTable.ConcatTable {
input
|`-> (0): nn.Identity
|`-> (1): nn.Sequential {
[input -> (0) -> (1) -> (2) -> (3) -> (4) -> (5) -> (6) -> output]
(0): nn.SpatialReplicationPadding(1, 1, 1, 1)
(1): nn.SpatialConvolution(128 -> 128, 3x3)
(2): nn.InstanceNormalization
(3): nn.ReLU
(4): nn.SpatialReplicationPadding(1, 1, 1, 1)
(5): nn.SpatialConvolution(128 -> 128, 3x3)
(6): nn.InstanceNormalization
}
+. -> output
}
(1): nn.CAddTable
}
(16): nn.SpatialFullConvolution(128 -> 64, 3x3, 2, 2, 1, 1, 1, 1)
(17): nn.InstanceNormalization
(18): nn.ReLU
(19): nn.SpatialFullConvolution(64 -> 32, 3x3, 2, 2, 1, 1, 1, 1)
(20): nn.InstanceNormalization
(21): nn.ReLU
(22): nn.SpatialReplicationPadding(1, 1, 1, 1)
(23): nn.SpatialConvolution(32 -> 3, 3x3)
}
the model is loaded perfectly fine but when i am trying to evaluate model , i get this error
Traceback (most recent call last):
File "convert-fast-neural-style.py", line 176, in <module>
main()
File "convert-fast-neural-style.py", line 162, in main
unknown_layer_converter_fn=convert_instance_norm
File "/usr/local/lib/python2.7/dist-packages/torch2coreml/_torch_converter.py", line 194, in convert
print (model.evaluate())
File "/usr/local/lib/python2.7/dist-packages/torch/legacy/nn/Container.py", line 39, in evaluate
self.applyToModules(lambda m: m.evaluate())
File "/usr/local/lib/python2.7/dist-packages/torch/legacy/nn/Container.py", line 26, in applyToModules
func(module)
File "/usr/local/lib/python2.7/dist-packages/torch/legacy/nn/Container.py", line 39, in <lambda>
self.applyToModules(lambda m: m.evaluate())
TypeError: 'NoneType' object is not callable
what is the main reason behind the evaluation to bring this NoneType error.
is their any other may that is equivalent to m.evaluate()
for reference also this is the Model i am trying to evaluate
The file convert-fast-neural-style.py does not belong to this repo, so I cannot help you...
i can provided for you but the part i am having an error with is is this
this is the load function
def load_torch_model(path):
model = load_lua(path, unknown_classes=True)
replace_module(
model,
lambda m: isinstance(m, TorchObject) and
m.torch_typename() == 'nn.InstanceNormalization',
create_instance_norm
)
replace_module(
model,
lambda m: isinstance(m, SpatialFullConvolution),
fix_full_conv
)
return model
this is the create
def convert_instance_norm(builder, name, layer, input_names, output_names):
if not isinstance(layer, InstanceNormalization):
raise TypeError('Unsupported type {}'.format(layer,))
epsilon = layer.eps
weight = layer.weight.numpy()
bias = None
if layer.bias is not None:
bias = layer.bias.numpy()
builder.add_batchnorm(
name=name,
channels=weight.shape[0],
gamma=weight,
beta=bias,
compute_mean_var=True,
instance_normalization=True,
input_name=input_names[0],
output_name=output_names[0],
epsilon=epsilon
)
return output_names
this is the full fix conv
def fix_full_conv(m):
m.finput = None
m.fgradInput = None
return m
when i use function evaluate i got the previous error , my final aim is to transfer this model to coreml
can you help ?
It seems you need to have instance norm defined in nn/Legacy in pytorch. I haven't ever tried to load torch models to pytorch, so I don't know how to do it.