yad2k.py 11 KB
#! /usr/bin/env python
"""
Reads Darknet53 config and weights and creates Keras model with TF backend.

Currently only supports layers in Darknet53 config.
"""

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, GlobalAveragePooling2D, Input, Reshape,
                          ZeroPadding2D, UpSampling2D, Activation, Lambda, MaxPooling2D)
from keras.layers.advanced_activations import LeakyReLU
from keras.layers.merge import concatenate, add
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='Yet Another 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(
    '-flcl',
    '--fully_convolutional',
    help='Model is fully convolutional so set input shape to (None, None, 3). '
    'WARNING: This experimental option does not work properly for YOLO_v2.',
    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')
    weights_header = np.ndarray(
        shape=(5, ), dtype='int32', buffer=weights_file.read(20))
    print('Weights Header: ', weights_header)
    # TODO: Check transpose flag when implementing fully connected layers.
    # transpose = (weight_header[0] > 1000) or (weight_header[1] > 1000)

    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.')
    if args.fully_convolutional:
        image_height, image_width = None, None
    else:
        image_height = int(cfg_parser['net_0']['height'])
        image_width = int(cfg_parser['net_0']['width'])

    prev_layer = Input(shape=(image_height, image_width, 3))
    all_layers = [prev_layer]
    outputs = []

    weight_decay = float(cfg_parser['net_0']['decay']
                         ) if 'net_0' in cfg_parser.sections() else 5e-4
    count = 0

    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]

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

            # TODO: This assumes channel last dim_ordering.
            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

                # TODO: Keras BatchNormalization mistakenly refers to var
                # as std.
                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)
            # TODO: Add check for Theano dim ordering.
            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))

            padding = 'same' if pad == 1 and stride == 1 else 'valid'
            # Adjust padding model for darknet.
            if stride == 2:
                prev_layer = ZeroPadding2D(((1, 0), (1, 0)))(prev_layer)

            # Create Conv2D 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(prev_layer)

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

        elif section.startswith('avgpool'):
            if cfg_parser.items(section) != []:
                raise ValueError('{} with params unsupported.'.format(section))
            all_layers.append(GlobalAveragePooling2D()(prev_layer))
            prev_layer = all_layers[-1]

        elif section.startswith('route'):
            ids = [int(i) for i in cfg_parser[section]['layers'].split(',')]
            if len(ids) == 2:
                for i, item in enumerate(ids):
                    if item != -1:
                        ids[i] = item + 1

            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('shortcut'):
            ids = [int(i) for i in cfg_parser[section]['from'].split(',')][0]
            activation = cfg_parser[section]['activation']
            shortcut = add([all_layers[ids], prev_layer])
            if activation == 'linear':
                shortcut = Activation('linear')(shortcut)
            all_layers.append(shortcut)
            prev_layer = all_layers[-1]

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

        elif section.startswith('yolo'):
            classes = int(cfg_parser[section]['classes'])
            # num = int(cfg_parser[section]['num'])
            # mask = int(cfg_parser[section]['mask'])
            n1, n2 = int(prev_layer.shape[1]), int(prev_layer.shape[2])
            n3 = 3
            n4 = (4 + 1 + classes)
            yolo = Reshape((n1, n2, n3, n4))(prev_layer)
            all_layers.append(yolo)
            prev_layer = all_layers[-1]
            outputs.append(len(all_layers) - 1)

        elif (section.startswith('net')):
            pass  # Configs not currently handled during model definition.
        else:
            raise ValueError(
                'Unsupported section header type: {}'.format(section))

    # Create and save model.
    model = Model(inputs=all_layers[0],
                  outputs=[all_layers[i] for i in outputs])
    print(model.summary())
    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))

    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())