jhasegaw/warp-ctc

PyTorch bindings for Warp-ctc

Why does this clone of https://github.com/jnishi/warp-ctc exist?

ESPnet downloads warp-ctc from github by default. jnishi's version assumes GPU and OpenMP, so it crashes the install process on my Mac, which has neither. So this version is designed for people like me who are debugging ESPnet code on a laptop that has no GPU code, and using Apple's brain-dead openMP-lacking compiler.

#############################

This is an extension onto the original repo found here.

Installation

Install PyTorch v0.4.

WARP_CTC_PATH should be set to the location of a built WarpCTC (i.e. libwarpctc.so). This defaults to ../build, so from within a new warp-ctc clone you could build WarpCTC like this:

git clone https://github.com/SeanNaren/warp-ctc.git
cd warp-ctc
mkdir build; cd build
cmake ..
make

Now install the bindings:

cd pytorch_binding
python setup.py install

If you try the above and get a dlopen error on OSX with anaconda3 (as recommended by pytorch):

cd ../pytorch_binding
python setup.py install
cd ../build
cp libwarpctc.dylib /Users/$WHOAMI/anaconda3/lib

This will resolve the library not loaded error. This can be easily modified to work with other python installs if needed.

Example to use the bindings below.

import torch
from warpctc_pytorch import CTCLoss
ctc_loss = CTCLoss()
# expected shape of seqLength x batchSize x alphabet_size
probs = torch.FloatTensor([[[0.1, 0.6, 0.1, 0.1, 0.1], [0.1, 0.1, 0.6, 0.1, 0.1]]]).transpose(0, 1).contiguous()
labels = torch.IntTensor([1, 2])
label_sizes = torch.IntTensor([2])
probs_sizes = torch.IntTensor([2])
probs.requires_grad_(True)  # tells autograd to compute gradients for probs
cost = ctc_loss(probs, labels, probs_sizes, label_sizes)
cost.backward()

Documentation

CTCLoss(size_average=False, length_average=False, reduce=True)
    # size_average (bool): normalize the loss by the batch size (default: False)
    # length_average (bool): normalize the loss by the total number of frames in the batch. If True, supersedes size_average (default: False)
    # reduce (bool): average or sum over observation for each minibatch.
        If `False`, returns a loss per batch element instead and ignores `average` options.
        (default: `True`)

forward(acts, labels, act_lens, label_lens)
    # acts: Tensor of (seqLength x batch x outputDim) containing output activations from network (before softmax)
    # labels: 1 dimensional Tensor containing all the targets of the batch in one large sequence
    # act_lens: Tensor of size (batch) containing size of each output sequence from the network
    # label_lens: Tensor of (batch) containing label length of each example