Initialize state

Insert quantization operations to the specified activation node. Args: * graph: TensorFlow graph * node_name: activation node's name * is_tra

Normalize with the given running averages of mean value & standard deviation. Args: * smpl_mat: unnormalized sample matrix * rms: running avera

get the height of a convolution output

get the width of a convolution output

Returns `slim.arg_scope` with all None arguments removed. Arguments: funcs: Functions to pass to `arg_scope`. **kwargs: Arguments to pass t

Returns the dimensions of an image tensor. Args: image: A rank-D Tensor. For 3-D of shape: `[height, width, channels]`. rank: The expected

Build the training / evaluation graph. Args: * is_train: whether to create the training graph

build a evaluation model from pruned model

Calculate the mask-pruned gradients. Args: * grads_origin: list of original gradients Returns: * grads_pruned: list of mask-pruned g

Calculate the mask-pruned gradients. Args: * grads_origin: list of original gradients Returns: * grads_pruned: list of mask-pruned g

Calculate the loss function's value and evaluation metrics. Returns: * loss: loss function's value * metrics: evaluation metrics

reshape weights according to bucket for 'channel' type. Note that for fc layers, buckets are created row-wisely. Args: w: A Tensor

Compute the convolution computation flops of the model

Get the left convolutions after pruning so far

Constructor function. Args: * scope: name scope in which the model is defined

Use quantiles of weights to initialize the quantization points. If bucketing is enabled, both 'split' and 'channel' applies. Args: *

Restore a model from the latest checkpoint files. Args: * is_train: whether to restore a model for training

Restore a model from the latest checkpoint files. Args: * is_train: whether to restore a model for training

Restore a model from the latest checkpoint files. Args: * is_train: whether to restore a model for training

Restore a model from the latest checkpoint files. Args: * is_train: whether to restore a model for training

Restore a model from the latest checkpoint files. Args: * is_train: whether to restore a model for training

Save the current model.

reshape weights according to bucket for 'split' type. Args: * w: A Tensor (weights) Returns: * A Tensor, shape [bucket_size, bucket_

Uniformly initialize the quantization points. If bucketing is enabled, both 'split' and 'channel' applies. Args: * nb_clusters: A Te

Uniform quantization function Args: * x: A Tensor (weights or activation output) * mbits: A scalar Tensor, tf.int64, spicifying number of

Calculate extra storage for the bucket scalling factors Args: * alpha: a Tensor, the scalling factor * beta: a Tensor, the scalling fact

Pads the input along the spatial dimensions independently of input size. Pads the input such that if it was used in a convolution with 'VALID' padd

Subtracts the given means from each image channel. For example: means = [123.68, 116.779, 103.939] image = _mean_image_subtraction(image, m

Process a complete data set and save it as a TFRecord. Args: name: string, unique identifier specifying the data set. directory: string, ro

Simple wrapper around tf.resize_images. This is primarily to make sure we use the same `ResizeMethod` and other details each time. Args: i

Sets arg scope defaults for all items present in defaults. Args: defaults: dictionary/list of pairs, containing a mapping from function to

Evenly splits num, num_ways so each piece is a multiple of divisible_by.

Unmap a subset of item (data) back to the original set of items (of size count)

Adjust the standard deviation to meet the distance requirement between actions. Args: * dst_curr: current distance between clean & distorted

Build a graph for training or evaluation. Args: * model_helper: model helper with definitions of model & dataset * unquant_node_names: list of

Calculate the weight-decaying loss. Args: * trainable_vars: list of trainable variables

Calculate the overall pruning ratio. Returns: * prune_ratio: overall pruning ratio

return the rl reward via reshaping

return the rl reward via reshaping

Convert *.pb model to a *.tflite model. Args: * net: network configurations * file_path_pb: file path to the *.pb model * file_path_tflite: f

create a pruner

Create a TensorFlow session. Return: * sess: TensorFlow session

Restore a model from the latest checkpoint files and then evaluate it.

Restore a model from the latest checkpoint files and then evaluate it.

Restore a model from the latest checkpoint files and then evaluate it.

Restore a model from the latest checkpoint files and then evaluate it.

evaluate the model

Export *.pb & *.tflite models from checkpoint files. Args: * net: network configurations * file_path_meta: file path to the *.meta data * fil

Forward pass function. Args: * inputs: inputs to the network's forward pass * is_train: whether to use the forward pass with training operation

Forward pass function. Args: * inputs: inputs to the network's forward pass * is_train: whether to use the forward pass with training operation

Forward pass function. Args: * inputs: input tensor to the network's forward pass * is_train: whether to use the forward pass with traini

Forward pass function. Args: * inputs: input tensor to the network's forward pass * is_train: whether to use the forward pass with training ope

Forward pass function. Args: * inputs: inputs to the network's forward pass * is_train: whether to use the forward pass with training operation

Forward pass function. Args: * inputs: inputs to the network's forward pass * data_format: data format ('channels_last' OR 'channels_first')

check whether the input operation is last layer before sum in a resual branch. Args: 'op': an operation Return: the name of

Get the data format of convolutional layers. Returns: * data_format: data format of convolutional layers

Get a list of maskable variables. Args: * trainable_vars: list of trainable variables Returns: * maskable_vars: list of maskable variables

Get all the operations of certain types in `otypes` within the model graph Args: otype: A list containing the type of returned operations

Get list of operations within certain name scope and also matches the pattern. Args: * scope: name scope * patterns: list of name patterns to b

Get operations for pertubing the model's parameters. Args: * model: original model * model_noisy: perturbed model * param_noise_std: standard

Get operations related to the target model. Args: * model: original model * model_tr: target model Returns: * init_op: initialization oper

Insert quantization operation for activation Args: * act_bit_dict: A dict with (key: act_op_name, value: act_bits)

Insert quantization operation for activation Args: * act_bit_dict: A dict with (key: act_op_name, value: act_bits)

Insert quantization operation for weights Args: * wewight_bit_dict: A dict with (key: matmul_op_name, value: quant_bits)

Insert quantization operation for weights Args: * wewight_bit_dict: A dict with (key: matmul_op_name, value: quant_bits)

Defines Mobilenet training scope. In default. We do not use BN ReWrite the scope.

get the number of channels of a convolution output

Convert a comma-separated list to a list of floating-point numbers.

filter out the variable_scope

filter out the variable_scope

calcualte relative error

Reset the standard deviation.

calculate the difference between before and after weight pruning for a certain branch sum

Restore a list of values from a plain text file. Args: * file_path: file path Returns: * vals: list of values

Sample a mini-batch of trasitions. Args: * batch_size: number of transitions in the mini-batch Returns: * mbatch: a mini-batch of tr