compressai_vision.model_wrappers

class compressai_vision.model_wrappers.BaseWrapper(device)

NOTE: virtual class to build your wrapper and interface with compressai_vision

An instance of this class helps you to wrap an off-the-shelf model so that the wrapped model can behave in various modes such as “full” and “partial” to process the input frames.

property cfg

deeper_features_for_accuracy_proxy(x: Dict)

compute accuracy proxy at the deeper layer than NN-Part1

features_to_output(x: Dict, device: str)

Complete the downstream task from the intermediate deep features

forward(x, input_map_function)

Complete the downstream task with end-to-end manner all the way from the input

input_to_features(x, device: str) → Dict

Computes deep features at the intermediate layer(s) all the way from the input

property model_cfg_path

property pretrained_weight_path

training: bool

class compressai_vision.model_wrappers.faster_rcnn_R_50_FPN_3x(device: str, **kwargs)

training: bool

class compressai_vision.model_wrappers.faster_rcnn_X_101_32x8d_FPN_3x(device: str, **kwargs)

training: bool

class compressai_vision.model_wrappers.jde_1088x608(device: str, **kwargs)

deeper_features_for_accuracy_proxy(x: Dict)

compute accuracy proxy at the deeper layer than NN-Part1

features_to_output(x: Dict, device: str)

Complete the downstream task from the intermediate deep features

forward(x)

Complete the downstream task with end-to-end manner all the way from the input

get_input_size(x)

Computes the size of the input image to the network

input_to_features(x, device: str) → Dict

Computes deep features at the intermediate layer(s) all the way from the input

reset()

training: bool

class compressai_vision.model_wrappers.mask_rcnn_R_50_FPN_3x(device: str, **kwargs)

training: bool

class compressai_vision.model_wrappers.mask_rcnn_X_101_32x8d_FPN_3x(device: str, **kwargs)

training: bool