gy15635282365
/
tensorlayer3

#! /usr/bin/python
# -*- coding: utf-8 -*-

import tensorflow as tf
import colorsys, random, cv2
import numpy as np
from tensorlayer.visualize import save_image


def decode_tf(conv_output, output_size, NUM_CLASS, STRIDES, ANCHORS, i=0, XYSCALE=[1, 1, 1]):
    batch_size = tf.shape(conv_output)[0]
    conv_output = tf.reshape(conv_output, (batch_size, output_size, output_size, 3, 5 + NUM_CLASS))

    conv_raw_dxdy, conv_raw_dwdh, conv_raw_conf, conv_raw_prob = tf.split(conv_output, (2, 2, 1, NUM_CLASS), axis=-1)

    xy_grid = tf.meshgrid(tf.range(output_size), tf.range(output_size))
    xy_grid = tf.expand_dims(tf.stack(xy_grid, axis=-1), axis=2)  # [gx, gy, 1, 2]
    xy_grid = tf.tile(tf.expand_dims(xy_grid, axis=0), [batch_size, 1, 1, 3, 1])

    xy_grid = tf.cast(xy_grid, tf.float32)

    pred_xy = ((tf.sigmoid(conv_raw_dxdy) * XYSCALE[i]) - 0.5 * (XYSCALE[i] - 1) + xy_grid) * \
              STRIDES[i]
    pred_wh = (tf.exp(conv_raw_dwdh) * ANCHORS[i])
    pred_xywh = tf.concat([pred_xy, pred_wh], axis=-1)

    pred_conf = tf.sigmoid(conv_raw_conf)
    pred_prob = tf.sigmoid(conv_raw_prob)

    pred_prob = pred_conf * pred_prob
    pred_prob = tf.reshape(pred_prob, (batch_size, -1, NUM_CLASS))
    pred_xywh = tf.reshape(pred_xywh, (batch_size, -1, 4))

    return pred_xywh, pred_prob


def decode(conv_output, output_size, NUM_CLASS, STRIDES, ANCHORS, i, XYSCALE=[1, 1, 1]):
    return decode_tf(conv_output, output_size, NUM_CLASS, STRIDES, ANCHORS, i=i, XYSCALE=XYSCALE)


def filter_boxes(box_xywh, scores, score_threshold=0.4, input_shape=tf.constant([416, 416])):
    scores_max = tf.math.reduce_max(scores, axis=-1)

    mask = scores_max >= score_threshold
    class_boxes = tf.boolean_mask(box_xywh, mask)
    pred_conf = tf.boolean_mask(scores, mask)
    class_boxes = tf.reshape(class_boxes, [tf.shape(scores)[0], -1, tf.shape(class_boxes)[-1]])
    pred_conf = tf.reshape(pred_conf, [tf.shape(scores)[0], -1, tf.shape(pred_conf)[-1]])

    box_xy, box_wh = tf.split(class_boxes, (2, 2), axis=-1)

    input_shape = tf.cast(input_shape, dtype=tf.float32)
    box_yx = box_xy[..., ::-1]
    box_hw = box_wh[..., ::-1]

    box_mins = (box_yx - (box_hw / 2.)) / input_shape
    box_maxes = (box_yx + (box_hw / 2.)) / input_shape
    boxes = tf.concat(
        [
            box_mins[..., 0:1],  # y_min
            box_mins[..., 1:2],  # x_min
            box_maxes[..., 0:1],  # y_max
            box_maxes[..., 1:2]  # x_max
        ],
        axis=-1
    )
    # return tf.concat([boxes, pred_conf], axis=-1)
    return (boxes, pred_conf)


def read_class_names(class_file_name):
    names = {}
    with open(class_file_name, 'r') as data:
        for ID, name in enumerate(data):
            names[ID] = name.strip('\n')
    return names


def draw_bbox(image, bboxes, show_label=True):
    classes = read_class_names('model/coco.names')
    num_classes = len(classes)
    image_h, image_w, _ = image.shape
    hsv_tuples = [(1.0 * x / num_classes, 1., 1.) for x in range(num_classes)]
    colors = list(map(lambda x: colorsys.hsv_to_rgb(*x), hsv_tuples))
    colors = list(map(lambda x: (int(x[0] * 255), int(x[1] * 255), int(x[2] * 255)), colors))

    random.seed(0)
    random.shuffle(colors)
    random.seed(None)

    out_boxes, out_scores, out_classes, num_boxes = bboxes
    for i in range(num_boxes[0]):
        if int(out_classes[0][i]) < 0 or int(out_classes[0][i]) > num_classes: continue
        coor = out_boxes[0][i]
        coor[0] = int(coor[0] * image_h)
        coor[2] = int(coor[2] * image_h)
        coor[1] = int(coor[1] * image_w)
        coor[3] = int(coor[3] * image_w)

        fontScale = 0.5
        score = out_scores[0][i]
        class_ind = int(out_classes[0][i])
        bbox_color = colors[class_ind]
        bbox_thick = int(0.6 * (image_h + image_w) / 600)
        c1, c2 = (coor[1], coor[0]), (coor[3], coor[2])
        cv2.rectangle(image, c1, c2, bbox_color, bbox_thick)

        if show_label:
            bbox_mess = '%s: %.2f' % (classes[class_ind], score)
            t_size = cv2.getTextSize(bbox_mess, 0, fontScale, thickness=bbox_thick // 2)[0]
            c3 = (c1[0] + t_size[0], c1[1] - t_size[1] - 3)
            cv2.rectangle(image, c1, (np.float32(c3[0]), np.float32(c3[1])), bbox_color, -1)  #filled

            cv2.putText(
                image, bbox_mess, (c1[0], np.float32(c1[1] - 2)), cv2.FONT_HERSHEY_SIMPLEX, fontScale, (0, 0, 0),
                bbox_thick // 2, lineType=cv2.LINE_AA
            )
    return image


def get_anchors(anchors_path, tiny=False):
    anchors = np.array(anchors_path)
    if tiny:
        return anchors.reshape(2, 3, 2)
    else:
        return anchors.reshape(3, 3, 2)


def decode_train(conv_output, output_size, NUM_CLASS, STRIDES, ANCHORS, i=0, XYSCALE=[1, 1, 1]):
    conv_output = tf.reshape(conv_output, (tf.shape(conv_output)[0], output_size, output_size, 3, 5 + NUM_CLASS))

    conv_raw_dxdy, conv_raw_dwdh, conv_raw_conf, conv_raw_prob = tf.split(conv_output, (2, 2, 1, NUM_CLASS), axis=-1)

    xy_grid = tf.meshgrid(tf.range(output_size), tf.range(output_size))
    xy_grid = tf.expand_dims(tf.stack(xy_grid, axis=-1), axis=2)  # [gx, gy, 1, 2]
    xy_grid = tf.tile(tf.expand_dims(xy_grid, axis=0), [tf.shape(conv_output)[0], 1, 1, 3, 1])

    xy_grid = tf.cast(xy_grid, tf.float32)

    pred_xy = ((tf.sigmoid(conv_raw_dxdy) * XYSCALE[i]) - 0.5 * (XYSCALE[i] - 1) + xy_grid) * \
              STRIDES[i]
    pred_wh = (tf.exp(conv_raw_dwdh) * ANCHORS[i])
    pred_xywh = tf.concat([pred_xy, pred_wh], axis=-1)

    pred_conf = tf.sigmoid(conv_raw_conf)
    pred_prob = tf.sigmoid(conv_raw_prob)

    return tf.concat([pred_xywh, pred_conf, pred_prob], axis=-1)


def yolo4_input_processing(original_image):
    image_data = cv2.resize(original_image, (416, 416))
    image_data = image_data / 255.
    images_data = []
    for i in range(1):
        images_data.append(image_data)
    images_data = np.asarray(images_data).astype(np.float32)
    batch_data = tf.constant(images_data)
    return batch_data


def yolo4_output_processing(feature_maps):
    STRIDES = [8, 16, 32]
    ANCHORS = get_anchors([12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401])
    NUM_CLASS = 80
    XYSCALE = [1.2, 1.1, 1.05]
    iou_threshold = 0.45
    score_threshold = 0.25

    bbox_tensors = []
    prob_tensors = []
    score_thres = 0.2
    for i, fm in enumerate(feature_maps):
        if i == 0:
            output_tensors = decode(fm, 416 // 8, NUM_CLASS, STRIDES, ANCHORS, i, XYSCALE)
        elif i == 1:
            output_tensors = decode(fm, 416 // 16, NUM_CLASS, STRIDES, ANCHORS, i, XYSCALE)
        else:
            output_tensors = decode(fm, 416 // 32, NUM_CLASS, STRIDES, ANCHORS, i, XYSCALE)
        bbox_tensors.append(output_tensors[0])
        prob_tensors.append(output_tensors[1])
    pred_bbox = tf.concat(bbox_tensors, axis=1)
    pred_prob = tf.concat(prob_tensors, axis=1)
    boxes, pred_conf = filter_boxes(
        pred_bbox, pred_prob, score_threshold=score_thres, input_shape=tf.constant([416, 416])
    )
    pred = {'concat': tf.concat([boxes, pred_conf], axis=-1)}

    for key, value in pred.items():
        boxes = value[:, :, 0:4]
        pred_conf = value[:, :, 4:]

    boxes, scores, classes, valid_detections = tf.image.combined_non_max_suppression(
        boxes=tf.reshape(boxes, (tf.shape(boxes)[0], -1, 1, 4)),
        scores=tf.reshape(pred_conf, (tf.shape(pred_conf)[0], -1, tf.shape(pred_conf)[-1])),
        max_output_size_per_class=50, max_total_size=50, iou_threshold=iou_threshold, score_threshold=score_threshold
    )
    output = [boxes.numpy(), scores.numpy(), classes.numpy(), valid_detections.numpy()]
    return output


def result_to_json(image, pred_bbox):
    image_h, image_w, _ = image.shape
    out_boxes, out_scores, out_classes, num_boxes = pred_bbox
    class_names = {}
    json_result = []
    with open('model/coco.names', 'r') as data:
        for ID, name in enumerate(data):
            class_names[ID] = name.strip('\n')
    nums_class = len(class_names)

    for i in range(num_boxes[0]):
        if int(out_classes[0][i]) < 0 or int(out_classes[0][i]) > nums_class: continue
        coor = out_boxes[0][i]
        coor[0] = int(coor[0] * image_h)
        coor[2] = int(coor[2] * image_h)
        coor[1] = int(coor[1] * image_w)
        coor[3] = int(coor[3] * image_w)

        score = float(out_scores[0][i])
        class_ind = int(out_classes[0][i])
        bbox = np.array([coor[1], coor[0], coor[3], coor[2]]).tolist()  # [x1,y1,x2,y2]
        json_result.append({'image': None, 'category_id': class_ind, 'bbox': bbox, 'score': score})

    return json_result


def draw_boxes_and_labels_to_image_with_json(image, json_result, class_list, save_name=None):
    """Draw bboxes and class labels on image. Return the image with bboxes.

    Parameters
    -----------
    image : numpy.array
        The RGB image [height, width, channel].
    json_result : list of dict
        The object detection result with json format.
    classes_list : list of str
        For converting ID to string on image.
    save_name : None or str
        The name of image file (i.e. image.png), if None, not to save image.

    Returns
    -------
    numpy.array
        The saved image.

    References
    -----------
    - OpenCV rectangle and putText.
    - `scikit-image <http://scikit-image.org/docs/dev/api/skimage.draw.html#skimage.draw.rectangle>`__.

    """
    image_h, image_w, _ = image.shape
    num_classes = len(class_list)
    hsv_tuples = [(1.0 * x / num_classes, 1., 1.) for x in range(num_classes)]
    colors = list(map(lambda x: colorsys.hsv_to_rgb(*x), hsv_tuples))
    colors = list(map(lambda x: (int(x[0] * 255), int(x[1] * 255), int(x[2] * 255)), colors))
    random.seed(0)
    random.shuffle(colors)
    random.seed(None)
    bbox_thick = int(0.6 * (image_h + image_w) / 600)
    fontScale = 0.5

    for bbox_info in json_result:
        image_name = bbox_info['image']
        category_id = bbox_info['category_id']
        if category_id < 0 or category_id > num_classes: continue
        bbox = bbox_info['bbox']  # the order of coordinates is [x1, y2, x2, y2]
        score = bbox_info['score']

        bbox_color = colors[category_id]
        c1, c2 = (int(bbox[0]), int(bbox[1])), (int(bbox[2]), int(bbox[3]))
        cv2.rectangle(image, c1, c2, bbox_color, bbox_thick)

        bbox_mess = '%s: %.2f' % (class_list[category_id], score)
        t_size = cv2.getTextSize(bbox_mess, 0, fontScale, thickness=bbox_thick // 2)[0]
        c3 = (c1[0] + t_size[0], c1[1] - t_size[1] - 3)
        cv2.rectangle(image, c1, (np.float32(c3[0]), np.float32(c3[1])), bbox_color, -1)

        cv2.putText(
            image, bbox_mess, (c1[0], np.float32(c1[1] - 2)), cv2.FONT_HERSHEY_SIMPLEX, fontScale, (0, 0, 0),
            bbox_thick // 2, lineType=cv2.LINE_AA
        )

    if save_name is not None:
        save_image(image, save_name)

    return image