add new dimensions at the front and the tail

scheduler in training stage

judge if crop face and compute crop radius

sum over the first and last dimention

Traverse the input module and its child recursively and replace all instance of torch.nn.modules.batchnorm.BatchNorm*N*d to Synchronized

Execute an replication callback `__data_parallel_replicate__` on each module created by original replication. The callback will be invoked w

load openface landmark from .csv file

Args: master_callback: a callback to be invoked after having collected messages from slave devices.

Compute the mean and standard-deviation with sum and square-sum. This method also maintains the moving average on the master device.

crop face according to openface landmark

extract audio files from videos

extract deep speech feature

extract video frames from videos

generate 3D coordinate grid

Register an slave device. Args: identifier: an identifier, usually is the device id. Returns: a `SlavePipe` obj

Main entry for the master device in each forward pass. The messages were first collected from each devices (including the master devi

Reduce the sum and square-sum, compute the statistics, and broadcast it.