Help-System

Installation

1. downloade the repository
2. create "\model" folder in the repository
3. download the network files from this link and place it in the "\model" folder. https://drive.google.com/drive/folders/1aSM6h17ZQYegrOA8IK_qw2eN_nF4rOWV?usp=sharing

Usage and functions

Here is a list of the usefull functions contained in the class. to use start by calling the class:

import activity_network
  
helpnet = activity_network.activity_network()

flow = helpnet.compute_optical_flow(frame, frame_prev)

returns the optical flow given two frame

pafMat, heatMat = helpnet.compute_pose(img)

returns openPose pafMat and heatMat

frame_preprocessed = helpnet.compound_channel(img, flow, heatMat, pafMat)

stack the rgb image, the optical flow, the heatMat and the pafMat along the channel dimension returning a 7 channel picture.

second_matrix = helpnet.compound_second_frames(frames)

groups into correct format a list "frames" of preprocessed frame. Temporaly ordered such that the first frame in the list is the oldest while the last is the newest.

input_tensor = helpnet.create_input_tensor_given_seconds(seconds)

create network input tensor given list of correctly formatted second matrices. Temporaly ordered such that the the first matrix in the list must be the oldest, the last the newest.

input_tensor = helpnet.create_input_tensor_given_preprocessed_frame(frames_collection)

create network input tensor given list of list of preprocessed frames. frames_collection must be a list of list of frames, where each list of frame contains the list of preprocessed frame extracted from one second. the frames must be ordered such that the first frame must be the oldest and the last the newest. The list of those list must follow the same order with the first list containing the frame of the oldest second and the last the one of the newest

now_softmax, next_softmax, help_softmax = helpnet.compute_activity_given_tensor(input_tensor, second_count)

returns the softmax for the now and next activity and the help action. It requires in iput the input tensor and the second count of the newest given second.

now_softmax, next_softmax, help_softmax = helpnet.compute_activity_given_seconds_matrix(input_tensor, second_count)

return the outputs given the list of seconds matrix

now_softmax, next_softmax, help_softmax = helpnet.compute_activity_given_frame_list(input_tensor, second_count)

return the outputs given the list of frames matrix