train.py

# -*- coding: utf-8 -*-
"""
Created on Fri Dec 21 18:44:46 2018

@author: wmy
"""

import numpy as np
import keras
import keras.backend as K
from keras.layers import Input, Lambda
from keras.models import Model
from keras.callbacks import TensorBoard, ModelCheckpoint, EarlyStopping

from model import preprocess_true_boxes, yolo_body, yolo_loss, get_random_data
    
train_path = 'info_gpu/train.txt'
val_path = 'info_gpu/val.txt'
test_path = 'info_gpu/test.txt'
save_path = 'my_gpu_model/'
class_path = 'infos/classes.txt'
anchor_path = 'infos/anchors.txt'

input_shape = (416,416)

def get_anchors(anchors_path):
    with open(anchors_path) as f:
        anchors = f.readline()
    anchors = [float(x) for x in anchors.split(',')]
    return np.array(anchors).reshape(-1, 2)

def get_classes(classes_path):
    with open(classes_path) as f:
        class_names = f.readlines()
    class_names = [c.strip() for c in class_names]
    return class_names

def create_model(input_shape, anchors, num_classes, load_pretrained=False, freeze_body=False,
            weights_path=save_path+'weights.h5'):
    K.clear_session() # get a new session
    image_input = Input(shape=(None, None, 3))
    h, w = input_shape
    num_anchors = len(anchors)
    y_true = [Input(shape=(h//{0:32, 1:16, 2:8}[l], w//{0:32, 1:16, 2:8}[l], \
        num_anchors//3, num_classes+5)) for l in range(3)]

    model_body = yolo_body(image_input, num_anchors//3, num_classes)
    print('Create YOLOv3 model with {} anchors and {} classes.'.format(num_anchors, num_classes))

    if load_pretrained:#如果载入权重，这里的权重就是原模型训练好的h5文件
        model_body.load_weights(weights_path, by_name=True, skip_mismatch=True)
        print('Load weights {}.'.format(weights_path))
        if freeze_body:
            # Do not freeze 3 output layers.
            num = len(model_body.layers)-7
            for i in range(num): model_body.layers[i].trainable = False
            print('Freeze the first {} layers of total {} layers.'.format(num, len(model_body.layers)))

    model_loss = Lambda(yolo_loss, output_shape=(1,), name='yolo_loss',
        arguments={'anchors': anchors, 'num_classes': num_classes, 'ignore_thresh': 0.5})(
        [*model_body.output, *y_true])
    model = Model([model_body.input, *y_true], model_loss)
    
    return model

def train(model, input_shape, anchors, num_classes, lr=0.001, epochs=20, log_dir=save_path):
    model.compile(optimizer=keras.optimizers.Adam(lr=lr, beta_1=0.9, beta_2=0.999, epsilon=1e-8), \
                  loss={'yolo_loss': lambda y_true, y_pred: y_pred})#编译模型
    logging = TensorBoard(log_dir=log_dir)
    checkpoint = ModelCheckpoint(log_dir + "ep{epoch:03d}-loss{loss:.3f}-val_loss{val_loss:.3f}.h5",
        monitor='val_loss', save_weights_only=True, save_best_only=True, period=1)
    batch_size = 2
    with open(train_path) as f:
        train = f.readlines()
    with open(val_path) as f:
        val = f.readlines()
    num_train = len(train)
    num_val = len(val)
    print('Train on {} samples, val on {} samples, with batch size {}.'.format(num_train, num_val, batch_size))

    model.fit_generator(data_generator_wrap(train, batch_size, input_shape, anchors, num_classes),
            steps_per_epoch=max(1, num_train//batch_size),
            validation_data=data_generator_wrap(val, batch_size, input_shape, anchors, num_classes),
            validation_steps=max(1, num_val//batch_size),
            epochs=epochs,
            initial_epoch=0)
    model.save_weights(log_dir + 'weights.h5')
    model.save(log_dir + 'model.h5')

#数据整理
def data_generator(annotation_lines, batch_size, input_shape, anchors, num_classes):
    n = len(annotation_lines)
    np.random.shuffle(annotation_lines)
    i = 0
    while True:
        image_data = []
        box_data = []
        for b in range(batch_size):
            i %= n
            image, box = get_random_data(annotation_lines[i], input_shape, random=True)
            image_data.append(image)
            box_data.append(box)
            i += 1
        image_data = np.array(image_data)
        box_data = np.array(box_data)
        y_true = preprocess_true_boxes(box_data, input_shape, anchors, num_classes)

        yield [image_data, *y_true], np.zeros(batch_size)

def data_generator_wrap(annotation_lines, batch_size, input_shape, anchors, num_classes):
    n = len(annotation_lines)
    if n==0 or batch_size<=0: 
        return None
    return data_generator(annotation_lines, batch_size, input_shape, anchors, num_classes)

if __name__ == '__main__':
    classes = get_classes(class_path)#得到类，其实只有一个类，所以为1
    anchors = get_anchors(anchor_path)#得到anchor数量
    model = create_model(input_shape, anchors, len(classes), load_pretrained=True)#获取模型，可能是加载也可以是重新生成
    train(model, input_shape, anchors, len(classes), lr=0.00075, epochs=20, log_dir=save_path)
    pass