keras h5—darknet weights模型相互转换

一.版本介绍

keras 2.1.5
tensorflow 1.13.1

这里以一个yolov3_tiny模型为案例。

二.保存方式

首先准备自己的h5模型；这里要注意h5模型的保存方式，不同保存方式，对应不同加载方式，如下：

> 训练:  model.save()    
> 加载:  model.load_model()

> 训练:  model.save_weights()    
> 加载:    
> 		model = LeNet()     
> 		model.load_weights(weight_path) 

三.转换过程

注：这里相互转换都需要提供模型的cfg文件。

• ① darknet to keras

import argparse
import configparser
import io
import os
from collections import defaultdict

import numpy as np
from keras import backend as K
from keras.layers import (Conv2D, Input, ZeroPadding2D, Add,
                          UpSampling2D, MaxPooling2D, Concatenate)
from keras.layers.advanced_activations import LeakyReLU
from keras.layers.normalization import BatchNormalization
from keras.models import Model
from keras.regularizers import l2
from keras.utils.vis_utils import plot_model as plot


parser = argparse.ArgumentParser(description='Darknet To Keras Converter.')
parser.add_argument('config_path', help='Path to Darknet cfg file.')
parser.add_argument('weights_path', help='Path to Darknet weights file.')
parser.add_argument('output_path', help='Path to output Keras model file.')
parser.add_argument(
    '-p',
    '--plot_model',
    help='Plot generated Keras model and save as image.',
    action='store_true')
parser.add_argument(
    '-w',
    '--weights_only',
    help='Save as Keras weights file instead of model file.',
    action='store_true')

def unique_config_sections(config_file):
    """Convert all config sections to have unique names.

    Adds unique suffixes to config sections for compability with configparser.
    """
    section_counters = defaultdict(int)
    output_stream = io.StringIO()
    with open(config_file) as fin:
        for line in fin:
            if line.startswith('['):
                section = line.strip().strip('[]')
                _section = section + '_' + str(section_counters[section])
                section_counters[section] += 1
                line = line.replace(section, _section)
            output_stream.write(line)
    output_stream.seek(0)
    return output_stream
    
def _main(args):
    config_path = os.path.expanduser(args.config_path)
    weights_path = os.path.expanduser(args.weights_path)
    assert config_path.endswith('.cfg'), '{} is not a .cfg file'.format(
        config_path)
    assert weights_path.endswith(
        '.weights'), '{} is not a .weights file'.format(weights_path)

    output_path = os.path.expanduser(args.output_path)
    assert output_path.endswith(
        '.h5'), 'output path {} is not a .h5 file'.format(output_path)
    output_root = os.path.splitext(output_path)[0]

    # Load weights and config.
    print('Loading weights.')
    weights_file = open(weights_path, 'rb')
    major, minor, revision = np.ndarray(
        shape=(3, ), dtype='int32', buffer=weights_file.read(12))
    if (major*10+minor)>=2 and major<1000 and minor<1000:
        seen = np.ndarray(shape=(1,), dtype='int64', buffer=weights_file.read(8))
    else:
        seen = np.ndarray(shape=(1,), dtype='int32', buffer=weights_file.read(4))
    print('Weights Header: ', major, minor, revision, seen)

    print('Parsing Darknet config.')
    unique_config_file = unique_config_sections(config_path)
    cfg_parser = configparser.ConfigParser()
    cfg_parser.read_file(unique_config_file)

    print('Creating Keras model.')
    input_layer = Input(shape=(None, None, 3))
    prev_layer = input_layer
    all_layers = []

    weight_decay = float(cfg_parser['net_0']['decay']
                         ) if 'net_0' in cfg_parser.sections() else 5e-4
    count = 0
    out_index = []
    for section in cfg_parser.sections():
        print('Parsing section {}'.format(section))
        if section.startswith('convolutional'):
            filters = int(cfg_parser[section]['filters'])
            size = int(cfg_parser[section]['size'])
            stride = int(cfg_parser[section]['stride'])
            pad = int(cfg_parser[section]['pad'])
            activation = cfg_parser[section]['activation']
            batch_normalize = 'batch_normalize' in cfg_parser[section]

            padding = 'same' if pad == 1 and stride == 1 else 'valid'

            # Setting weights.
            # Darknet serializes convolutional weights as:
            # [bias/beta, [gamma, mean, variance], conv_weights]
            prev_layer_shape = K.int_shape(prev_layer)

            weights_shape = (size, size, prev_layer_shape[-1], filters)
            darknet_w_shape = (filters, weights_shape[2], size, size)
            weights_size = np.product(weights_shape)

            print('conv2d', 'bn'
                  if batch_normalize else '  ', activation, weights_shape)

            conv_bias = np.ndarray(
                shape=(filters, ),
                dtype='float32',
                buffer=weights_file.read(filters * 4))
            count += filters

            if batch_normalize:
                bn_weights = np.ndarray(
                    shape=(3, filters),
                    dtype='float32',
                    buffer=weights_file.read(filters * 12))
                count += 3 * filters

                bn_weight_list = [
                    bn_weights[0],  # scale gamma
                    conv_bias,  # shift beta
                    bn_weights[1],  # running mean
                    bn_weights[2]  # running var
                ]

            conv_weights = np.ndarray(
                shape=darknet_w_shape,
                dtype='float32',
                buffer=weights_file.read(weights_size * 4))
            count += weights_size

            # DarkNet conv_weights are serialized Caffe-style:
            # (out_dim, in_dim, height, width)
            # We would like to set these to Tensorflow order:
            # (height, width, in_dim, out_dim)
            conv_weights = np.transpose(conv_weights, [2, 3, 1, 0])
            conv_weights = [conv_weights] if batch_normalize else [
                conv_weights, conv_bias
            ]

            # Handle activation.
            act_fn = None
            if activation == 'leaky':
                pass  # Add advanced activation later.
            elif activation != 'linear':
                raise ValueError(
                    'Unknown activation function `{}` in section {}'.format(
                        activation, section))

            # Create Conv2D layer
            if stride>1:
                # Darknet uses left and top padding instead of 'same' mode
                prev_layer = ZeroPadding2D(((1,0),(1,0)))(prev_layer)
            conv_layer = (Conv2D(
                filters, (size, size),
                strides=(stride, stride),
                kernel_regularizer=l2(weight_decay),
                use_bias=not batch_normalize,
                weights=conv_weights,
                activation=act_fn,
                padding=padding))(prev_layer)

            if batch_normalize:
                conv_layer = (BatchNormalization(
                    weights=bn_weight_list))(conv_layer)
            prev_layer = conv_layer

            if activation == 'linear':
                all_layers.append(prev_layer)
            elif activation == 'leaky':
                act_layer = LeakyReLU(alpha=0.1)(prev_layer)
                prev_layer = act_layer
                all_layers.append(act_layer)

        elif section.startswith('route'):
            ids = [int(i) for i in cfg_parser[section]['layers'].split(',')]
            layers = [all_layers[i] for i in ids]
            if len(layers) > 1:
                print('Concatenating route layers:', layers)
                concatenate_layer = Concatenate()(layers)
                all_layers.append(concatenate_layer)
                prev_layer = concatenate_layer
            else:
                skip_layer = layers[0]  # only one layer to route
                all_layers.append(skip_layer)
                prev_layer = skip_layer

        elif section.startswith('maxpool'):
            size = int(cfg_parser[section]['size'])
            stride = int(cfg_parser[section]['stride'])
            all_layers.append(
                MaxPooling2D(
                    pool_size=(size, size),
                    strides=(stride, stride),
                    padding='same')(prev_layer))
            prev_layer = all_layers[-1]

        elif section.startswith('shortcut'):
            index = int(cfg_parser[section]['from'])
            activation = cfg_parser[section]['activation']
            assert activation == 'linear', 'Only linear activation supported.'
            all_layers.append(Add()([all_layers[index], prev_layer]))
            prev_layer = all_layers[-1]

        elif section.startswith('upsample'):
            stride = int(cfg_parser[section]['stride'])
            assert stride == 2, 'Only stride=2 supported.'
            all_layers.append(UpSampling2D(stride)(prev_layer))
            prev_layer = all_layers[-1]

        elif section.startswith('yolo'):
            out_index.append(len(all_layers)-1)
            all_layers.append(None)
            prev_layer = all_layers[-1]

        elif section.startswith('net'):
            pass

        else:
            raise ValueError(
                'Unsupported section header type: {}'.format(section))

    # Create and save model.
    if len(out_index)==0: out_index.append(len(all_layers)-1)
    model = Model(inputs=input_layer, outputs=[all_layers[i] for i in out_index])
    print(model.summary())
    if args.weights_only:
        model.save_weights('{}'.format(output_path))
        print('Saved Keras weights to {}'.format(output_path))
    else:
        model.save('{}'.format(output_path))
        print('Saved Keras model to {}'.format(output_path))

    # Check to see if all weights have been read.
    remaining_weights = len(weights_file.read()) / 4
    weights_file.close()
    print('Read {} of {} from Darknet weights.'.format(count, count +
                                                       remaining_weights))
    if remaining_weights > 0:
        print('Warning: {} unused weights'.format(remaining_weights))

    if args.plot_model:
        plot(model, to_file='{}.png'.format(output_root), show_shapes=True)
        print('Saved model plot to {}.png'.format(output_root))

if __name__ == '__main__':
    _main(parser.parse_args())

• ② keras to darknet
  根据保存方式判断是否需要model.py

"""model.py"""

from functools import wraps

import numpy as np
import tensorflow as tf
from keras import backend as K
from keras.layers import Conv2D, Add, ZeroPadding2D, UpSampling2D, Concatenate, MaxPooling2D
from keras.layers.advanced_activations import LeakyReLU
from keras.layers.normalization import BatchNormalization
from keras.models import Model
from keras.regularizers import l2

from yolo3.utils import compose


@wraps(Conv2D)
def DarknetConv2D(*args, **kwargs):
    """Wrapper to set Darknet parameters for Convolution2D."""
    darknet_conv_kwargs = {'kernel_regularizer': l2(5e-4)}
    darknet_conv_kwargs['padding'] = 'valid' if kwargs.get('strides')==(2,2) else 'same'
    darknet_conv_kwargs.update(kwargs)
    return Conv2D(*args, **darknet_conv_kwargs)

def DarknetConv2D_BN_Leaky(*args, **kwargs):
    """Darknet Convolution2D followed by BatchNormalization and LeakyReLU."""
    no_bias_kwargs = {'use_bias': False}
    no_bias_kwargs.update(kwargs)
    return compose(
        DarknetConv2D(*args, **no_bias_kwargs),
        BatchNormalization(),
        LeakyReLU(alpha=0.1))

def resblock_body(x, num_filters, num_blocks):
    '''A series of resblocks starting with a downsampling Convolution2D'''
    # Darknet uses left and top padding instead of 'same' mode
    x = ZeroPadding2D(((1,0),(1,0)))(x)
    x = DarknetConv2D_BN_Leaky(num_filters, (3,3), strides=(2,2))(x)
    for i in range(num_blocks):
        y = compose(
                DarknetConv2D_BN_Leaky(num_filters//2, (1,1)),
                DarknetConv2D_BN_Leaky(num_filters, (3,3)))(x)
        x = Add()([x,y])
    return x

def darknet_body(x):
    '''Darknent body having 52 Convolution2D layers'''
    x = DarknetConv2D_BN_Leaky(32, (3,3))(x)
    x = resblock_body(x, 64, 1)
    x = resblock_body(x, 128, 2)
    x = resblock_body(x, 256, 8)
    x = resblock_body(x, 512, 8)
    x = resblock_body(x, 1024, 4)
    return x

def make_last_layers(x, num_filters, out_filters):
    '''6 Conv2D_BN_Leaky layers followed by a Conv2D_linear layer'''
    x = compose(
            DarknetConv2D_BN_Leaky(num_filters, (1,1)),
            DarknetConv2D_BN_Leaky(num_filters*2, (3,3)),
            DarknetConv2D_BN_Leaky(num_filters, (1,1)),
            DarknetConv2D_BN_Leaky(num_filters*2, (3,3)),
            DarknetConv2D_BN_Leaky(num_filters, (1,1)))(x)
    y = compose(
            DarknetConv2D_BN_Leaky(num_filters*2, (3,3)),
            DarknetConv2D(out_filters, (1,1)))(x)
    return x, y


def yolo_body(inputs, num_anchors, num_classes):
    """Create YOLO_V3 model CNN body in Keras."""
    darknet = Model(inputs, darknet_body(inputs))
    x, y1 = make_last_layers(darknet.output, 512, num_anchors*(num_classes+5))

    x = compose(
            DarknetConv2D_BN_Leaky(256, (1,1)),
            UpSampling2D(2))(x)
    x = Concatenate()([x,darknet.layers[152].output])
    x, y2 = make_last_layers(x, 256, num_anchors*(num_classes+5))

    x = compose(
            DarknetConv2D_BN_Leaky(128, (1,1)),
            UpSampling2D(2))(x)
    x = Concatenate()([x,darknet.layers[92].output])
    x, y3 = make_last_layers(x, 128, num_anchors*(num_classes+5))

    return Model(inputs, [y1,y2,y3])

def tiny_yolo_body(inputs, num_anchors, num_classes):
    '''Create Tiny YOLO_v3 model CNN body in keras.'''
    x1 = compose(
            DarknetConv2D_BN_Leaky(16, (3,3)),
            MaxPooling2D(pool_size=(2,2), strides=(2,2), padding='same'),
            DarknetConv2D_BN_Leaky(32, (3,3)),
            MaxPooling2D(pool_size=(2,2), strides=(2,2), padding='same'),
            DarknetConv2D_BN_Leaky(64, (3,3)),
            MaxPooling2D(pool_size=(2,2), strides=(2,2), padding='same'),
            DarknetConv2D_BN_Leaky(128, (3,3)),
            MaxPooling2D(pool_size=(2,2), strides=(2,2), padding='same'),
            DarknetConv2D_BN_Leaky(256, (3,3)))(inputs)
    x2 = compose(
            MaxPooling2D(pool_size=(2,2), strides=(2,2), padding='same'),
            DarknetConv2D_BN_Leaky(512, (3,3)),
            MaxPooling2D(pool_size=(2,2), strides=(1,1), padding='same'),
            DarknetConv2D_BN_Leaky(1024, (3,3)),
            DarknetConv2D_BN_Leaky(256, (1,1)))(x1)
    y1 = compose(
            DarknetConv2D_BN_Leaky(512, (3,3)),
            DarknetConv2D(num_anchors*(num_classes+5), (1,1)))(x2)

    x2 = compose(
            DarknetConv2D_BN_Leaky(128, (1,1)),
            UpSampling2D(2))(x2)
    y2 = compose(
            Concatenate(),
            DarknetConv2D_BN_Leaky(256, (3,3)),
            DarknetConv2D(num_anchors*(num_classes+5), (1,1)))([x2,x1])

    return Model(inputs, [y1,y2])


def yolo_head(feats, anchors, num_classes, input_shape, calc_loss=False):
    """Convert final layer features to bounding box parameters."""
    num_anchors = len(anchors)
    # Reshape to batch, height, width, num_anchors, box_params.
    anchors_tensor = K.reshape(K.constant(anchors), [1, 1, 1, num_anchors, 2])

    grid_shape = K.shape(feats)[1:3] # height, width
    grid_y = K.tile(K.reshape(K.arange(0, stop=grid_shape[0]), [-1, 1, 1, 1]),
        [1, grid_shape[1], 1, 1])
    grid_x = K.tile(K.reshape(K.arange(0, stop=grid_shape[1]), [1, -1, 1, 1]),
        [grid_shape[0], 1, 1, 1])
    grid = K.concatenate([grid_x, grid_y])
    grid = K.cast(grid, K.dtype(feats))

    feats = K.reshape(
        feats, [-1, grid_shape[0], grid_shape[1], num_anchors, num_classes + 5])

    # Adjust preditions to each spatial grid point and anchor size.
    box_xy = (K.sigmoid(feats[..., :2]) + grid) / K.cast(grid_shape[::-1], K.dtype(feats))
    box_wh = K.exp(feats[..., 2:4]) * anchors_tensor / K.cast(input_shape[::-1], K.dtype(feats))
    box_confidence = K.sigmoid(feats[..., 4:5])
    box_class_probs = K.sigmoid(feats[..., 5:])

    if calc_loss == True:
        return grid, feats, box_xy, box_wh
    return box_xy, box_wh, box_confidence, box_class_probs


def yolo_correct_boxes(box_xy, box_wh, input_shape, image_shape):
    '''Get corrected boxes'''
    box_yx = box_xy[..., ::-1]
    box_hw = box_wh[..., ::-1]
    input_shape = K.cast(input_shape, K.dtype(box_yx))
    image_shape = K.cast(image_shape, K.dtype(box_yx))
    new_shape = K.round(image_shape * K.min(input_shape/image_shape))
    offset = (input_shape-new_shape)/2./input_shape
    scale = input_shape/new_shape
    box_yx = (box_yx - offset) * scale
    box_hw *= scale

    box_mins = box_yx - (box_hw / 2.)
    box_maxes = box_yx + (box_hw / 2.)
    boxes =  K.concatenate([
        box_mins[..., 0:1],  # y_min
        box_mins[..., 1:2],  # x_min
        box_maxes[..., 0:1],  # y_max
        box_maxes[..., 1:2]  # x_max
    ])

    # Scale boxes back to original image shape.
    boxes *= K.concatenate([image_shape, image_shape])
    return boxes


def yolo_boxes_and_scores(feats, anchors, num_classes, input_shape, image_shape):
    '''Process Conv layer output'''
    box_xy, box_wh, box_confidence, box_class_probs = yolo_head(feats,
        anchors, num_classes, input_shape)
    boxes = yolo_correct_boxes(box_xy, box_wh, input_shape, image_shape)
    boxes = K.reshape(boxes, [-1, 4])
    box_scores = box_confidence * box_class_probs
    box_scores = K.reshape(box_scores, [-1, num_classes])
    return boxes, box_scores


def yolo_eval(yolo_outputs,
              anchors,
              num_classes,
              image_shape,
              max_boxes=20,
              score_threshold=.6,
              iou_threshold=.5):
    """Evaluate YOLO model on given input and return filtered boxes."""
    num_layers = len(yolo_outputs)
    anchor_mask = [[6,7,8], [3,4,5], [0,1,2]] if num_layers==3 else [[3,4,5], [1,2,3]] # default setting
    input_shape = K.shape(yolo_outputs[0])[1:3] * 32
    boxes = []
    box_scores = []
    for l in range(num_layers):
        _boxes, _box_scores = yolo_boxes_and_scores(yolo_outputs[l],
            anchors[anchor_mask[l]], num_classes, input_shape, image_shape)
        boxes.append(_boxes)
        box_scores.append(_box_scores)
    boxes = K.concatenate(boxes, axis=0)
    box_scores = K.concatenate(box_scores, axis=0)

    mask = box_scores >= score_threshold
    max_boxes_tensor = K.constant(max_boxes, dtype='int32')
    boxes_ = []
    scores_ = []
    classes_ = []
    for c in range(num_classes):
        # TODO: use keras backend instead of tf.
        class_boxes = tf.boolean_mask(boxes, mask[:, c])
        class_box_scores = tf.boolean_mask(box_scores[:, c], mask[:, c])
        nms_index = tf.image.non_max_suppression(
            class_boxes, class_box_scores, max_boxes_tensor, iou_threshold=iou_threshold)
        class_boxes = K.gather(class_boxes, nms_index)
        class_box_scores = K.gather(class_box_scores, nms_index)
        classes = K.ones_like(class_box_scores, 'int32') * c
        boxes_.append(class_boxes)
        scores_.append(class_box_scores)
        classes_.append(classes)
    boxes_ = K.concatenate(boxes_, axis=0)
    scores_ = K.concatenate(scores_, axis=0)
    classes_ = K.concatenate(classes_, axis=0)

    return boxes_, scores_, classes_


def preprocess_true_boxes(true_boxes, input_shape, anchors, num_classes):
    '''Preprocess true boxes to training input format

    Parameters
    ----------
    true_boxes: array, shape=(m, T, 5)
        Absolute x_min, y_min, x_max, y_max, class_id relative to input_shape.
    input_shape: array-like, hw, multiples of 32
    anchors: array, shape=(N, 2), wh
    num_classes: integer

    Returns
    -------
    y_true: list of array, shape like yolo_outputs, xywh are reletive value

    '''
    assert (true_boxes[..., 4]<num_classes).all(), 'class id must be less than num_classes'
    num_layers = len(anchors)//3 # default setting
    anchor_mask = [[6,7,8], [3,4,5], [0,1,2]] if num_layers==3 else [[3,4,5], [1,2,3]]

    true_boxes = np.array(true_boxes, dtype='float32')
    input_shape = np.array(input_shape, dtype='int32')
    boxes_xy = (true_boxes[..., 0:2] + true_boxes[..., 2:4]) // 2
    boxes_wh = true_boxes[..., 2:4] - true_boxes[..., 0:2]
    true_boxes[..., 0:2] = boxes_xy/input_shape[::-1]
    true_boxes[..., 2:4] = boxes_wh/input_shape[::-1]

    m = true_boxes.shape[0]
    grid_shapes = [input_shape//{0:32, 1:16, 2:8}[l] for l in range(num_layers)]
    y_true = [np.zeros((m,grid_shapes[l][0],grid_shapes[l][1],len(anchor_mask[l]),5+num_classes),
        dtype='float32') for l in range(num_layers)]

    # Expand dim to apply broadcasting.
    anchors = np.expand_dims(anchors, 0)
    anchor_maxes = anchors / 2.
    anchor_mins = -anchor_maxes
    valid_mask = boxes_wh[..., 0]>0

    for b in range(m):
        # Discard zero rows.
        wh = boxes_wh[b, valid_mask[b]]
        if len(wh)==0: continue
        # Expand dim to apply broadcasting.
        wh = np.expand_dims(wh, -2)
        box_maxes = wh / 2.
        box_mins = -box_maxes

        intersect_mins = np.maximum(box_mins, anchor_mins)
        intersect_maxes = np.minimum(box_maxes, anchor_maxes)
        intersect_wh = np.maximum(intersect_maxes - intersect_mins, 0.)
        intersect_area = intersect_wh[..., 0] * intersect_wh[..., 1]
        box_area = wh[..., 0] * wh[..., 1]
        anchor_area = anchors[..., 0] * anchors[..., 1]
        iou = intersect_area / (box_area + anchor_area - intersect_area)

        # Find best anchor for each true box
        best_anchor = np.argmax(iou, axis=-1)

        for t, n in enumerate(best_anchor):
            for l in range(num_layers):
                if n in anchor_mask[l]:
                    i = np.floor(true_boxes[b,t,0]*grid_shapes[l][1]).astype('int32')
                    j = np.floor(true_boxes[b,t,1]*grid_shapes[l][0]).astype('int32')
                    k = anchor_mask[l].index(n)
                    c = true_boxes[b,t, 4].astype('int32')
                    y_true[l][b, j, i, k, 0:4] = true_boxes[b,t, 0:4]
                    y_true[l][b, j, i, k, 4] = 1
                    y_true[l][b, j, i, k, 5+c] = 1

    return y_true


def box_iou(b1, b2):
    '''Return iou tensor

    Parameters
    ----------
    b1: tensor, shape=(i1,...,iN, 4), xywh
    b2: tensor, shape=(j, 4), xywh

    Returns
    -------
    iou: tensor, shape=(i1,...,iN, j)

    '''

    # Expand dim to apply broadcasting.
    b1 = K.expand_dims(b1, -2)
    b1_xy = b1[..., :2]
    b1_wh = b1[..., 2:4]
    b1_wh_half = b1_wh/2.
    b1_mins = b1_xy - b1_wh_half
    b1_maxes = b1_xy + b1_wh_half

    # Expand dim to apply broadcasting.
    b2 = K.expand_dims(b2, 0)
    b2_xy = b2[..., :2]
    b2_wh = b2[..., 2:4]
    b2_wh_half = b2_wh/2.
    b2_mins = b2_xy - b2_wh_half
    b2_maxes = b2_xy + b2_wh_half

    intersect_mins = K.maximum(b1_mins, b2_mins)
    intersect_maxes = K.minimum(b1_maxes, b2_maxes)
    intersect_wh = K.maximum(intersect_maxes - intersect_mins, 0.)
    intersect_area = intersect_wh[..., 0] * intersect_wh[..., 1]
    b1_area = b1_wh[..., 0] * b1_wh[..., 1]
    b2_area = b2_wh[..., 0] * b2_wh[..., 1]
    iou = intersect_area / (b1_area + b2_area - intersect_area)

    return iou


def yolo_loss(args, anchors, num_classes, ignore_thresh=.5, print_loss=False):
    '''Return yolo_loss tensor

    Parameters
    ----------
    yolo_outputs: list of tensor, the output of yolo_body or tiny_yolo_body
    y_true: list of array, the output of preprocess_true_boxes
    anchors: array, shape=(N, 2), wh
    num_classes: integer
    ignore_thresh: float, the iou threshold whether to ignore object confidence loss

    Returns
    -------
    loss: tensor, shape=(1,)

    '''
    num_layers = len(anchors)//3 # default setting
    yolo_outputs = args[:num_layers]
    y_true = args[num_layers:]
    anchor_mask = [[6,7,8], [3,4,5], [0,1,2]] if num_layers==3 else [[3,4,5], [1,2,3]]
    input_shape = K.cast(K.shape(yolo_outputs[0])[1:3] * 32, K.dtype(y_true[0]))
    grid_shapes = [K.cast(K.shape(yolo_outputs[l])[1:3], K.dtype(y_true[0])) for l in range(num_layers)]
    loss = 0
    m = K.shape(yolo_outputs[0])[0] # batch size, tensor
    mf = K.cast(m, K.dtype(yolo_outputs[0]))

    for l in range(num_layers):
        object_mask = y_true[l][..., 4:5]
        true_class_probs = y_true[l][..., 5:]

        grid, raw_pred, pred_xy, pred_wh = yolo_head(yolo_outputs[l],
             anchors[anchor_mask[l]], num_classes, input_shape, calc_loss=True)
        pred_box = K.concatenate([pred_xy, pred_wh])

        # Darknet raw box to calculate loss.
        raw_true_xy = y_true[l][..., :2]*grid_shapes[l][::-1] - grid
        raw_true_wh = K.log(y_true[l][..., 2:4] / anchors[anchor_mask[l]] * input_shape[::-1])
        raw_true_wh = K.switch(object_mask, raw_true_wh, K.zeros_like(raw_true_wh)) # avoid log(0)=-inf
        box_loss_scale = 2 - y_true[l][...,2:3]*y_true[l][...,3:4]

        # Find ignore mask, iterate over each of batch.
        ignore_mask = tf.TensorArray(K.dtype(y_true[0]), size=1, dynamic_size=True)
        object_mask_bool = K.cast(object_mask, 'bool')
        def loop_body(b, ignore_mask):
            true_box = tf.boolean_mask(y_true[l][b,...,0:4], object_mask_bool[b,...,0])
            iou = box_iou(pred_box[b], true_box)
            best_iou = K.max(iou, axis=-1)
            ignore_mask = ignore_mask.write(b, K.cast(best_iou<ignore_thresh, K.dtype(true_box)))
            return b+1, ignore_mask
        _, ignore_mask = K.control_flow_ops.while_loop(lambda b,*args: b<m, loop_body, [0, ignore_mask])
        ignore_mask = ignore_mask.stack()
        ignore_mask = K.expand_dims(ignore_mask, -1)

        # K.binary_crossentropy is helpful to avoid exp overflow.
        xy_loss = object_mask * box_loss_scale * K.binary_crossentropy(raw_true_xy, raw_pred[...,0:2], from_logits=True)
        wh_loss = object_mask * box_loss_scale * 0.5 * K.square(raw_true_wh-raw_pred[...,2:4])
        confidence_loss = object_mask * K.binary_crossentropy(object_mask, raw_pred[...,4:5], from_logits=True)+ \
            (1-object_mask) * K.binary_crossentropy(object_mask, raw_pred[...,4:5], from_logits=True) * ignore_mask
        class_loss = object_mask * K.binary_crossentropy(true_class_probs, raw_pred[...,5:], from_logits=True)

        xy_loss = K.sum(xy_loss) / mf
        wh_loss = K.sum(wh_loss) / mf
        confidence_loss = K.sum(confidence_loss) / mf
        class_loss = K.sum(class_loss) / mf
        loss += xy_loss + wh_loss + confidence_loss + class_loss
        if print_loss:
            loss = tf.Print(loss, [loss, xy_loss, wh_loss, confidence_loss, class_loss, K.sum(ignore_mask)], message='loss: ')
    return loss

转换脚本：

# -*- coding: utf-8 -*-
import argparse
import numpy
import numpy as np
from keras.layers import Input
from keras.models import load_model
from keras import backend as K
from yolo3.model import tiny_yolo_body

def parser():
    parser = argparse.ArgumentParser(
        description="Darknet\'s yolov3.cfg and yolov3.weights converted into Keras\'s yolov3.h5!")
    parser.add_argument('-cfg_path', default='yolov3.cfg',help='yolov3.cfg')
    parser.add_argument('-h5_path', default='mask.h5',help='A.h5')
    parser.add_argument('-output_path',default='model_data/VVVVVV.weights',help='yolov3.weights')
    return parser.parse_args()


class WeightSaver(object):

    def __init__(self, h5_path, output_path):
        try:
            self.model = load_model(h5_path, compile=False)
        except:
            self.model = tiny_yolo_body(Input(shape=(None, None,3)), 6 // 2, 2)
            self.model.load_weights(h5_path)

        self.layers = {weight.name: weight for weight in self.model.weights}
        self.sess = K.get_session()
        self.fhandle = open(output_path, 'wb')
        self._write_head()

    def _write_head(self):
        numpy_data = numpy.ndarray(shape=(3,),
                                   dtype='int32',
                                   buffer=np.array([0, 2, 0], dtype='int32'))
        self.save(numpy_data)
        numpy_data = numpy.ndarray(shape=(1,),
                                   dtype='int64',
                                   buffer=np.array([320000], dtype='int64'))
        self.save(numpy_data)

    def get_bn_layername(self, num):
        layer_name = f'batch_normalization_{num}'
        bias = self.layers[f'{layer_name}/beta:0']
        scale = self.layers[f'{layer_name}/gamma:0']
        mean = self.layers[f'{layer_name}/moving_mean:0']
        var = self.layers[f'{layer_name}/moving_variance:0']

        bias_np = self.get_numpy(bias)
        scale_np = self.get_numpy(scale)
        mean_np = self.get_numpy(mean)
        var_np = self.get_numpy(var)
        return bias_np, scale_np, mean_np, var_np

    def get_convbias_layername(self, num):
        layer_name = f'conv2d_{num}'
        bias = self.layers[f'{layer_name}/bias:0']

        bias_np = self.get_numpy(bias)
        return bias_np

    def get_conv_layername(self, num):
        layer_name = f'conv2d_{num}'
        conv = self.layers[f'{layer_name}/kernel:0']

        conv_np = self.get_numpy(conv)
        return conv_np

    def get_numpy(self, layer_name):
        numpy_data = self.sess.run(layer_name)
        return numpy_data

    def save(self, numpy_data):
        bytes_data = numpy_data.tobytes()
        self.fhandle.write(bytes_data)
        self.fhandle.flush()

    def close(self):
        self.fhandle.close()


class KerasParser(object):

    def __init__(self, cfg_path, h5_path, output_path):
        self.block_gen = self._get_block(cfg_path)
        self.weights_saver = WeightSaver(h5_path, output_path)
        self.count_conv = 0
        self.count_bn = 0

    def _get_block(self, cfg_path):

        block = {}
        with open(cfg_path, 'r', encoding='utf-8') as fr:
            for line in fr:
                line = line.strip()
                if '[' in line and ']' in line:
                    if block:
                        yield block
                    block = {}
                    block['type'] = line.strip(' []')
                elif not line or '#' in line:
                    continue
                else:
                    key, val = line.strip().replace(' ', '').split('=')
                    key, val = key.strip(), val.strip()
                    block[key] = val

            yield block

    def close(self):
        self.weights_saver.close()

    def conv(self, block):
        self.count_conv += 1
        batch_normalize = 'batch_normalize' in block

        # 如果bn存在，则先处理bn，顺序为bias，scale，mean，var
        if batch_normalize:
            bias, scale, mean, var = self.bn()
            self.weights_saver.save(bias)

            scale = scale.reshape(1, -1)
            mean = mean.reshape(1, -1)
            var = var.reshape(1, -1)
            remain = np.concatenate([scale, mean, var], axis=0)
            self.weights_saver.save(remain)

        # 否则，先处理biase
        else:
            conv_bias = self.weights_saver.get_convbias_layername(self.count_conv)
            self.weights_saver.save(conv_bias)

        # 接着处理weights
        conv_weights = self.weights_saver.get_conv_layername(self.count_conv)
        # 需要将(height, width, in_dim, out_dim)转换成(out_dim, in_dim, height, width)
        conv_weights = np.transpose(conv_weights, [3, 2, 0, 1])
        self.weights_saver.save(conv_weights)

    def bn(self):
        self.count_bn += 1
        bias, scale, mean, var = self.weights_saver.get_bn_layername(self.count_bn)
        return bias, scale, mean, var


def main():
    args = parser()
    keras_loader = KerasParser(args.cfg_path, args.h5_path, args.output_path)

    for block in keras_loader.block_gen:
        print(block)
        if 'convolutional' in block['type']:
            keras_loader.conv(block)
    keras_loader.close()

if __name__ == "__main__":
    main()
#使用方式
# python h5_to_weights.py -cfg_path ./model_data/yolov3.cfg -h5_path ./model_data/A.h5 -output_path ./model_data/A.weights

四.参考

https://github.com/qqwweee/keras-yolo3