utils.old.py

import numpy as np
from PIL import Image
import tensorflow as tf


def read_jpeg_image(img_path):
    image = tf.io.read_file(img_path)
    image = tf.image.decode_jpeg(image, channels=3)
    return image


def resize(image, min_side=800.0, max_side=1333.0):
    h = tf.cast(tf.shape(image)[0], tf.float32)
    w = tf.cast(tf.shape(image)[1], tf.float32)
    cur_min_side = tf.minimum(w, h)
    cur_max_side = tf.maximum(w, h)

    scale = tf.minimum(max_side / cur_max_side,
                       min_side / cur_min_side)
    nh = tf.cast(scale * h, tf.int32)
    nw = tf.cast(scale * w, tf.int32)

    image = tf.image.resize(image, (nh, nw))
    return image


def build_mask(image):
    h = tf.shape(image)[0]
    w = tf.shape(image)[1]
    return tf.zeros((h, w), dtype=tf.bool)


def cxcywh2xyxy(boxes):
    cx, cy, w, h = [boxes[..., i] for i in range(4)]

    xmin, ymin = cx - w*0.5, cy - h*0.5
    xmax, ymax = cx + w*0.5, cy + h*0.5

    boxes = tf.stack([xmin, ymin, xmax, ymax], axis=-1)
    return boxes


def absolute2relative(boxes, img_size):
    width, height = img_size
    scale = tf.constant([width, height, width, height], dtype=tf.float32)
    boxes *= scale
    return boxes


def xyxy2xywh(boxes):
    xmin, ymin, xmax, ymax = [boxes[..., i] for i in range(4)]
    return tf.stack([xmin, ymin, xmax - xmin, ymax - ymin], axis=-1)


def preprocess_image(image):
    image = resize(image, min_side=800.0, max_side=1333.0)

    channel_avg = tf.constant([0.485, 0.456, 0.406])
    channel_std = tf.constant([0.229, 0.224, 0.225])
    image = (image / 255.0 - channel_avg) / channel_std

    return image, build_mask(image)