halfish / lstm-ctc-ocr Goto Github PK

I get this output when I tried to run 3_phonernn.lua:

train size = 8, valid size = 2 nn.Sequential { [input -> (1) -> (2) -> (3) -> (4) -> (5) -> (6) -> (7) -> (8) -> output] (1): nn.SplitTable (2): nn.Sequencer @ nn.FastLSTM(36 -> 128) (3): nn.Sequencer @ nn.Recursor @ nn.ReLU (4): nn.Sequencer @ nn.Recursor @ nn.BatchNormalization (5): nn.Sequencer @ nn.Recursor @ nn.Dropout(0.5, busy) (6): nn.Sequencer @ nn.Recursor @ nn.Linear(128 -> 12) (7): nn.JoinTable (8): nn.View(255, 12) } [=================== 1/1 =====================>] Tot: 0ms | Step: 0ms /home/ubuntu/torch/install/bin/luajit: /home/ubuntu/torch/install/share/lua/5.1/nn/Container.lua:67: In 8 module of nn.Sequential: /home/ubuntu/torch/install/share/lua/5.1/torch/Tensor.lua:458: bad argument #1 to 'set' (expecting number or torch.DoubleTensor or torch.DoubleStorage at /home/ubuntu/torch/pkg/torch/generic/Tensor.c:1125)

What's happening here? How do I solve this?
I've run 1_generateImage.py and 2_dump.lua successfully prior to this. Thanks in advance.

@Halfish the images of the numeral-sequences you create are all of the same dimensions (36 x 255). How can one go about using variable width images instead? (36 x var-width)

I see that you feed in a complete image at a time to the model (not as 255-strips of 36pix-height of the image -- a sequence of pixel-strips). Can you suggest how this can be modified to incorporate variable length inputs?