sali_encode.py

'''
MIT License

Copyright (c) [2020] [Duin BAEK]

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
'''

import os
import cv2
import numpy as np
import pandas as pd
import multiprocessing
from saliency_update import *

def rotate_img(img, iteration = 1):
    #0: remain same, +n: rotate left in 90*n degree, -n: rotate right in 90*n degree (some part of img is sliced off)
    if len(img.shape) == 3:
        height, width, _ = img.shape
    else:
        height, width = img.shape

    rotated_img = np.zeros(img.shape, dtype = np.uint8)

    if iteration == -1:

        for idx in range(height):
            rotated_img[:, height-1-idx] = img[idx, :]

    elif iteration == 1:

        for idx in range(height):
            rotated_img[:, idx] = np.flip(img[idx, :], axis = 0)

    elif iteration == 0:

        rotated_img = img

    else:

        for idx in range(height):
            rotated_img[height-1-idx, :] = np.flip(img[idx, :], axis = 0)

    return rotated_img

def frame_split(frame):
    #return a list of screens in an order of right, left, up, down, front and back
    if len(frame.shape) == 3:
        height, width, _ = frame.shape
    else:
        height, width = frame.shape

    unit_height = int(height / 2)
    unit_width = int(width / 3)

    #1st layer
    right_frame = frame[:unit_height, :unit_width]
    left_frame = frame[:unit_height, unit_width:2*unit_width]
    up_frame = frame[:unit_height, 2*unit_width:]
    #2nd layer
    down_frame = frame[unit_height:, :unit_width]
    front_frame = frame[unit_height:, unit_width:2*unit_width]
    back_frame = frame[unit_height:, 2*unit_width:]

    return [right_frame, left_frame, up_frame, down_frame, front_frame, back_frame]

def trapezoid_1(img): # starting from small frame / 2 + 2

    if len(img.shape) == 3:
        row, col, channel = img.shape

    else:
        row, col = img.shape
        channel = 1

    empty_pallete = np.zeros((row, col, channel), dtype = np.uint8)

    for idx in range(int(row / 4)):
        layer = cv2.resize(img[idx*4:(idx+1)*4, :], (int(row/2) + 2 + 2*idx,1))
        _, length, _ = layer.shape

        empty_pallete[idx, int(row/4)-1-idx: int(row/4)-1-idx + length, :  ] = layer

        #cv2.imshow('original', img[idx*4:(idx+1)*4, :])
        #cv2.imshow('resized', layer)

    #cv2.imshow('pallete', empty_pallete)
    #cv2.waitKey(0)

    return empty_pallete #shape of row, col, channel (1024, 1024, 3)

def trapezoid_2(img): # starting from small frame / 2

    if len(img.shape) == 3:
        row, col, channel = img.shape

    else:
        row, col = img.shape
        channel = 1

    empty_pallete = np.zeros((row, col, channel), dtype = np.uint8)

    for idx in range(int(row / 4)):
        layer = cv2.resize(img[idx*4:(idx+1)*4, :], (int(row/2) + 2*idx,1))
        _, length, _ = layer.shape

        empty_pallete[idx, int(row/4)-idx: int(row/4)-idx + length, :  ] = layer

    return empty_pallete #shape of row, col, channel (1024, 1024, 3)

def pyramid_b_encoding(img):

    #[right_frame, left_frame, up_frame, down_frame, front_frame, back_frame]
    split_img = frame_split(img)

    row, col, channel = split_img[0].shape
    pallete = np.zeros((row, col, channel), dtype = np.uint8)

    #right: rotate left once and trapezoid_1 and rotate right, indexing from the right
    #left: rotate right once and trapezoid_1 and rotate left, indexing from the left
    #up: rotate none and trapezoid_2 rotate twice, indexing from the bottom
    #down: rotate twice and trapezoid_2, indexing from the top

    right_img = split_img[0]
    right_img = rotate_img(right_img, iteration = 1)

    right_trapezoid = trapezoid_1(right_img)
    right_trapezoid = rotate_img(right_trapezoid, iteration = -1)

    left_img = split_img[1]
    left_img = rotate_img(left_img, iteration = -1)

    left_trapezoid = trapezoid_1(left_img)
    left_trapezoid = rotate_img(left_trapezoid, iteration = 1)

    up_img = split_img[2]

    up_trapezoid = trapezoid_2(up_img)
    up_trapezoid = rotate_img(up_trapezoid, iteration = 2)


    down_img = split_img[3]

    down_img = rotate_img(down_img, iteration = 2)
    down_trapezoid = trapezoid_2(down_img)

    front_img = split_img[4]

    back_img = split_img[5]
    back_img = cv2.resize(back_img, None, fx = 0.5, fy = 0.5)

    height, width, _ = back_img.shape

    pallete[:, :int(col/4), :] = np.maximum(pallete[:, :int(col/4), :], right_trapezoid[:, -int(col/4):, :]) #right
    pallete[:, -int(col/4):, :] = np.maximum(pallete[:, -int(col/4):, :], left_trapezoid[:, :int(col/4), :]) #left
    pallete[:int(row/4), :, :] = np.maximum(pallete[:int(row/4), :, :], up_trapezoid[-int(row/4):, :, :]) #up
    pallete[-int(row/4):, :, :] = np.maximum(pallete[-int(row/4):, :, :], down_trapezoid[:int(row/4), :, :]) #down
    pallete[int(row/4):int(row/4)+height, int(col/4):int(col/4)+width, :] = back_img

    pyra_c_frame = np.concatenate([front_img, pallete], axis = 1)

    return pyra_c_frame

def frame_split_1(frame):
    #return a list of screens in an order of right, left, up, down, front and back
    if len(frame.shape) == 3:
        height, width, _ = frame.shape
    else:
        height, width = frame.shape

    unit_height = int(height / 2)
    unit_width = int(width / 3)

    #1st layer
    right_frame = frame[:unit_height, :unit_width]
    left_frame = frame[:unit_height, unit_width:2*unit_width]
    up_frame = frame[:unit_height, 2*unit_width:]
    #2nd layer
    down_frame = frame[unit_height:, :unit_width]
    front_frame = frame[unit_height:, unit_width:2*unit_width]
    back_frame = frame[unit_height:, 2*unit_width:]

    return [right_frame, left_frame, up_frame, down_frame, front_frame, back_frame]

def json_sort_by_name(y_p_combo_path):
    json_file_list = np.array(os.listdir(y_p_combo_path))
    idx_list = [int(json_file.split('.')[0].split('_')[-1]) for json_file in json_file_list]
    sorted_idx_list = np.argsort(idx_list)
    sorted_json_file_list = json_file_list[sorted_idx_list]

    return sorted_json_file_list

def json_path_generate(path, j2f_ratio):

    json_path_list = []

    sorted_json_file_list = json_sort_by_name(path)

    for json_file in sorted_json_file_list:
        for _ in range(j2f_ratio):
            json_path_list.append(os.path.join(path, json_file))

    return np.array(json_path_list).ravel()

def saliency_patching(cube_frame, json_path, frame_idx, fps, duration_unit, salient_patch_size, num_col):

    frame_height, frame_width, _ = cube_frame.shape
    side_length = int(frame_width / 3)
    window_size = int(side_length / salient_patch_size) #in the paper, salient_region_rate is set to 4

    name2idx = np.array(['R', 'L', 'U', 'D', 'F', 'B'])

    split_img = frame_split_1(cube_frame) #[right_frame, left_frame, up_frame, down_frame, front_frame, back_frame]

    json_data = saliency_score_update(json_path, frame_idx, fps, duration_unit)
    json_keys = np.array(list(json_data.keys()))

    sorted_json_idx = np.argsort([int(key) for key in json_keys])
    json_keys = json_keys[sorted_json_idx]

    img_cluster =[]
    layer_img = []
    threshold = 0
    #newly added
    num_row = 0
    '''#should change tar_row to change the number of salient patches to be used.'''
    tar_row = salient_patch_size

    for key in json_keys:
        img = split_img[np.where(name2idx == json_data[key]['name'])[0][0]][int(json_data[key]['row']): int(json_data[key]['row']) + window_size, int(json_data[key]['column']): int(json_data[key]['column']) + int(json_data[key]['width'])]

        layer_img.append(img)
        threshold += int(json_data[key]['width'])
        #print(threshold)

        if threshold == num_col*window_size:
            img_cluster.append(np.concatenate(layer_img, axis = 1))
            threshold = 0
            layer_img = []
            num_row += 1

        #print(num_row)
        if num_row == tar_row:
            break

    out_img = np.concatenate(img_cluster, axis = 0)
    return out_img

def encode(cube_frame, json_path, frame_idx, fps, duration_unit, salient_patch_size, num_col):

    pyra = pyramid_b_encoding(cube_frame)
    sali = saliency_patching(cube_frame, json_path, frame_idx, fps, duration_unit, salient_patch_size, num_col)

    result = np.concatenate([pyra, sali], axis = 1)
    return result

num_col = 2
j2f_ratio = 10
salient_patch_size = 4

cube_base_path = 'video/segments/cube'
json_base_path = 'json'
write_base_path = 'video/segments/sali_encoded'

fourcc = cv2.VideoWriter_fourcc(*'X264')

for y_p_combo in os.listdir(cube_base_path):

    y_p_combo_path = os.path.join(cube_base_path, y_p_combo)
    y_p_json_path = os.path.join(json_base_path, y_p_combo)
    y_p_result_path = os.path.join(write_base_path, y_p_combo)


    for duration in os.listdir(y_p_combo_path):

        duration_path = os.path.join(y_p_combo_path, duration)
        duration_json_path = os.path.join(y_p_json_path, duration)
        duration_result_path = os.path.join(y_p_result_path, duration)
        if not os.path.exists(duration_result_path):
            os.makedirs(duration_result_path)

        for file_name in sorted(os.listdir(duration_path)):

            file_path = os.path.join(duration_path, file_name)
            result_path = os.path.join(duration_result_path, file_name)
            json_path = json_path_generate(duration_json_path, j2f_ratio)

            capture = cv2.VideoCapture(file_path)

            fps = round(capture.get(cv2.CAP_PROP_FPS))
            frame_idx = 0

            while(capture.isOpened()):

                retval, frame = capture.read()

                if retval == True:

                    encoded = encode(frame, json_path[frame_idx], frame_idx, fps, duration, salient_patch_size, num_col)

                    if frame_idx == 0:
                        frame_height, frame_width, _ = encoded.shape
                        writer = cv2.VideoWriter(result_path, fourcc, fps, (frame_width, frame_height))

                    writer.write(encoded)

                    frame_idx += 1
                else:
                    break

            capture.release()
            writer.release()










#end