|
- import os
-
- import numpy as np
- import tensorflow as tf
- import tensorflow.keras.backend as K
- from tensorflow.keras.callbacks import (EarlyStopping, ReduceLROnPlateau,
- TensorBoard)
- from tensorflow.keras.layers import Input, Lambda
- from tensorflow.keras.models import Model
- from tensorflow.keras.optimizers import Adam
-
- from nets.loss import yolo_loss
- from nets.yolo4_tiny import yolo_body
- from utils.utils import (ModelCheckpoint, WarmUpCosineDecayScheduler,
- get_random_data, get_random_data_with_Mosaic, rand)
-
-
- #---------------------------------------------------#
- # 获得类和先验框
- #---------------------------------------------------#
- def get_classes(classes_path):
- '''loads the classes'''
- with open(classes_path) as f:
- class_names = f.readlines()
- class_names = [c.strip() for c in class_names]
- return class_names
-
- def get_anchors(anchors_path):
- '''loads the anchors from a file'''
- with open(anchors_path) as f:
- anchors = f.readline()
- anchors = [float(x) for x in anchors.split(',')]
- return np.array(anchors).reshape(-1, 2)
-
- #---------------------------------------------------#
- # 训练数据生成器
- #---------------------------------------------------#
- def data_generator(annotation_lines, batch_size, input_shape, anchors, num_classes, mosaic=False, random=True):
- n = len(annotation_lines)
- i = 0
- flag = True
- while True:
- image_data = []
- box_data = []
- for b in range(batch_size):
- if i==0:
- np.random.shuffle(annotation_lines)
- if mosaic:
- if flag and (i+4) < n:
- image, box = get_random_data_with_Mosaic(annotation_lines[i:i+4], input_shape)
- i = (i+4) % n
- else:
- image, box = get_random_data(annotation_lines[i], input_shape, random=random)
- i = (i+1) % n
- flag = bool(1-flag)
- else:
- image, box = get_random_data(annotation_lines[i], input_shape, random=random)
- i = (i+1) % n
- image_data.append(image)
- box_data.append(box)
- 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)
-
- #---------------------------------------------------#
- # 读入xml文件,并输出y_true
- #---------------------------------------------------#
- def preprocess_true_boxes(true_boxes, input_shape, anchors, num_classes):
- assert (true_boxes[..., 4]<num_classes).all(), 'class id must be less than num_classes'
- # 一共有两个特征层数
- num_layers = len(anchors)//3
- #-----------------------------------------------------------#
- # 13x13的特征层对应的anchor是[81,82], [135,169], [344,319]
- # 26x26的特征层对应的anchor是[23,27], [37,58], [81,82]
- #-----------------------------------------------------------#
- anchor_mask = [[6,7,8], [3,4,5], [0,1,2]] if num_layers==3 else [[3,4,5], [1,2,3]]
-
- #-----------------------------------------------------------#
- # 获得框的坐标和图片的大小
- #-----------------------------------------------------------#
- true_boxes = np.array(true_boxes, dtype='float32')
- input_shape = np.array(input_shape, dtype='int32')
- #-----------------------------------------------------------#
- # 通过计算获得真实框的中心和宽高
- # 中心点(m,n,2) 宽高(m,n,2)
- #-----------------------------------------------------------#
- boxes_xy = (true_boxes[..., 0:2] + true_boxes[..., 2:4]) // 2
- boxes_wh = true_boxes[..., 2:4] - true_boxes[..., 0:2]
- #-----------------------------------------------------------#
- # 将真实框归一化到小数形式
- #-----------------------------------------------------------#
- true_boxes[..., 0:2] = boxes_xy/input_shape[::-1]
- true_boxes[..., 2:4] = boxes_wh/input_shape[::-1]
-
- # m为图片数量,grid_shapes为网格的shape
- m = true_boxes.shape[0]
- grid_shapes = [input_shape//{0:32, 1:16, 2:8}[l] for l in range(num_layers)]
- #-----------------------------------------------------------#
- # y_true的格式为(m,13,13,3,85)(m,26,26,3,85)
- #-----------------------------------------------------------#
- 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)]
-
- #-----------------------------------------------------------#
- # [6,2] -> [1,6,2]
- #-----------------------------------------------------------#
- anchors = np.expand_dims(anchors, 0)
- anchor_maxes = anchors / 2.
- anchor_mins = -anchor_maxes
-
- #-----------------------------------------------------------#
- # 长宽要大于0才有效
- #-----------------------------------------------------------#
- valid_mask = boxes_wh[..., 0]>0
-
- for b in range(m):
- # 对每一张图进行处理
- wh = boxes_wh[b, valid_mask[b]]
- if len(wh)==0: continue
- #-----------------------------------------------------------#
- # [n,2] -> [n,1,2]
- #-----------------------------------------------------------#
- wh = np.expand_dims(wh, -2)
- box_maxes = wh / 2.
- box_mins = -box_maxes
-
- #-----------------------------------------------------------#
- # 计算所有真实框和先验框的交并比
- # intersect_area [n,6]
- # box_area [n,1]
- # anchor_area [1,6]
- # iou [n,6]
- #-----------------------------------------------------------#
- 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)
- #-----------------------------------------------------------#
- # 维度是[n,] 感谢 消尽不死鸟 的提醒
- #-----------------------------------------------------------#
- best_anchor = np.argmax(iou, axis=-1)
-
- for t, n in enumerate(best_anchor):
- #-----------------------------------------------------------#
- # 找到每个真实框所属的特征层
- #-----------------------------------------------------------#
- for l in range(num_layers):
- if n in anchor_mask[l]:
- #-----------------------------------------------------------#
- # floor用于向下取整,找到真实框所属的特征层对应的x、y轴坐标
- #-----------------------------------------------------------#
- 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指的的当前这个特征点的第k个先验框
- #-----------------------------------------------------------#
- k = anchor_mask[l].index(n)
- #-----------------------------------------------------------#
- # c指的是当前这个真实框的种类
- #-----------------------------------------------------------#
- c = true_boxes[b, t, 4].astype('int32')
- #-----------------------------------------------------------#
- # y_true的shape为(m,13,13,3,85)(m,26,26,3,85)(m,52,52,3,85)
- # 最后的85可以拆分成4+1+80,4代表的是框的中心与宽高、
- # 1代表的是置信度、80代表的是种类
- #-----------------------------------------------------------#
- 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
-
-
- gpus = tf.config.experimental.list_physical_devices(device_type='GPU')
- for gpu in gpus:
- tf.config.experimental.set_memory_growth(gpu, True)
-
-
- #----------------------------------------------------#
- # 检测精度mAP和pr曲线计算参考视频
- # https://www.bilibili.com/video/BV1zE411u7Vw
- #----------------------------------------------------#
- if __name__ == "__main__":
- #----------------------------------------------------#
- # 获得图片路径和标签
- #----------------------------------------------------#
- annotation_path = 'mnist_train.txt'
- #------------------------------------------------------#
- # 训练后的模型保存的位置,保存在logs文件夹里面
- #------------------------------------------------------#
- log_dir = 'logs/'
- #----------------------------------------------------#
- # classes和anchor的路径,非常重要
- # 训练前一定要修改classes_path,使其对应自己的数据集
- #----------------------------------------------------#
- classes_path = 'model_data/voc_classes.txt'
- anchors_path = 'model_data/yolo_anchors.txt'
- #------------------------------------------------------#
- # 权值文件请看README,百度网盘下载
- # 训练自己的数据集时提示维度不匹配正常
- # 预测的东西都不一样了自然维度不匹配
- #------------------------------------------------------#
- weights_path = 'model_data/yolov4_tiny_weights_voc.h5'
- #------------------------------------------------------#
- # 训练用图片大小
- # 一般在416x416和608x608选择
- #------------------------------------------------------#
- input_shape = (416,416)
- #------------------------------------------------------#
- # 是否对损失进行归一化
- #------------------------------------------------------#
- normalize = True
-
- #----------------------------------------------------#
- # 获取classes和anchor
- #----------------------------------------------------#
- class_names = get_classes(classes_path)
- anchors = get_anchors(anchors_path)
- #------------------------------------------------------#
- # 一共有多少类和多少先验框
- #------------------------------------------------------#
- num_classes = len(class_names)
- num_anchors = len(anchors)
- #------------------------------------------------------#
- # Yolov4的tricks应用
- # mosaic 马赛克数据增强 True or False
- # Cosine_scheduler 余弦退火学习率 True or False
- # label_smoothing 标签平滑 0.01以下一般 如0.01、0.005
- #------------------------------------------------------#
- mosaic = False
- Cosine_scheduler = False
- label_smoothing = 0
-
- #------------------------------------------------------#
- # 创建yolo模型
- #------------------------------------------------------#
- image_input = Input(shape=(None, None, 3))
- h, w = input_shape
- print('Create YOLOv4-Tiny model with {} anchors and {} classes.'.format(num_anchors, num_classes))
- model_body = yolo_body(image_input, num_anchors//2, num_classes)
-
- #-------------------------------------------#
- # 权值文件的下载请看README
- #-------------------------------------------#
- print('Load weights {}.'.format(weights_path))
- model_body.load_weights(weights_path, by_name=True, skip_mismatch=True)
-
- # y_true为13,13,3,85
- # 26,26,3,85
- y_true = [Input(shape=(h//{0:32, 1:16}[l], w//{0:32, 1:16}[l], num_anchors//2, num_classes+5)) for l in range(2)]
-
- #------------------------------------------------------#
- # 在这个地方设置损失,将网络的输出结果传入loss函数
- # 把整个模型的输出作为loss
- #------------------------------------------------------#
- loss_input = [*model_body.output, *y_true]
- model_loss = Lambda(yolo_loss, output_shape=(1,), name='yolo_loss',
- arguments={'anchors': anchors, 'num_classes': num_classes, 'ignore_thresh': 0.5, 'label_smoothing': label_smoothing})(loss_input)
- model = Model([model_body.input, *y_true], model_loss)
-
- #-------------------------------------------------------------------------------#
- # 训练参数的设置
- # logging表示tensorboard的保存地址
- # checkpoint用于设置权值保存的细节,period用于修改多少epoch保存一次
- # reduce_lr用于设置学习率下降的方式
- # early_stopping用于设定早停,val_loss多次不下降自动结束训练,表示模型基本收敛
- #-------------------------------------------------------------------------------#
- 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=False, period=1)
- early_stopping = EarlyStopping(monitor='val_loss', min_delta=0, patience=10, verbose=1)
-
- #----------------------------------------------------------------------#
- # 验证集的划分在train.py代码里面进行
- # 2007_test.txt和2007_val.txt里面没有内容是正常的。训练不会使用到。
- # 当前划分方式下,验证集和训练集的比例为1:9
- #----------------------------------------------------------------------#
- val_split = 0.125
- with open(annotation_path) as f:
- lines = f.readlines()
- np.random.seed(10101)
- np.random.shuffle(lines)
- np.random.seed(None)
- num_val = int(len(lines)*val_split)
- num_train = len(lines) - num_val
-
- freeze_layers = 60
- for i in range(freeze_layers): model_body.layers[i].trainable = False
- print('Freeze the first {} layers of total {} layers.'.format(freeze_layers, len(model_body.layers)))
-
- #------------------------------------------------------#
- # 主干特征提取网络特征通用,冻结训练可以加快训练速度
- # 也可以在训练初期防止权值被破坏。
- # Init_Epoch为起始世代
- # Freeze_Epoch为冻结训练的世代
- # Epoch总训练世代
- # 提示OOM或者显存不足请调小Batch_size
- #------------------------------------------------------#
- if True:
- Init_epoch = 0
- Freeze_epoch = 25
- batch_size = 64
- learning_rate_base = 1e-3
-
- if Cosine_scheduler:
- # 预热期
- warmup_epoch = int((Freeze_epoch-Init_epoch)*0.2)
- print(warmup_epoch)
- # 总共的步长
- total_steps = int((Freeze_epoch-Init_epoch) * num_train / batch_size)
- # 预热步长
- warmup_steps = int(warmup_epoch * num_train / batch_size)
- # 学习率
- reduce_lr = WarmUpCosineDecayScheduler(learning_rate_base=learning_rate_base,
- total_steps=total_steps,
- warmup_learning_rate=1e-4,
- warmup_steps=warmup_steps,
- hold_base_rate_steps=num_train,
- min_learn_rate=1e-6
- )
- model.compile(optimizer=Adam(), loss={'yolo_loss': lambda y_true, y_pred: y_pred})
- else:
- reduce_lr = ReduceLROnPlateau(monitor='val_loss', factor=0.5, patience=3, verbose=1)
- model.compile(optimizer=Adam(learning_rate_base), loss={'yolo_loss': lambda y_true, y_pred: y_pred})
-
- print('Train on {} samples, val on {} samples, with batch size {}.'.format(num_train, num_val, batch_size))
- model.fit(data_generator(lines[:num_train], batch_size, input_shape, anchors, num_classes, mosaic=mosaic),
- steps_per_epoch=max(1, num_train//batch_size),
- validation_data=data_generator(lines[num_train:], batch_size, input_shape, anchors, num_classes, mosaic=False),
- validation_steps=max(1, num_val//batch_size),
- epochs=Freeze_epoch,
- initial_epoch=Init_epoch,
- callbacks=[logging, checkpoint, reduce_lr, early_stopping])
- model.save_weights(log_dir + 'trained_weights_stage_1.h5')
-
- for i in range(freeze_layers): model_body.layers[i].trainable = True
-
- # 解冻后训练
- if True:
- print("解冻")
- Freeze_epoch = 25
- Epoch = 50
- batch_size = 64
- learning_rate_base = 1e-4
-
- if Cosine_scheduler:
- # 预热期
- warmup_epoch = int((Epoch-Freeze_epoch)*0.2)
- print(warmup_epoch)
- # 总共的步长
- total_steps = int((Epoch-Freeze_epoch) * num_train / batch_size)
- # 预热步长
- warmup_steps = int(warmup_epoch * num_train / batch_size)
- # 学习率
- reduce_lr = WarmUpCosineDecayScheduler(learning_rate_base=learning_rate_base,
- total_steps=total_steps,
- warmup_learning_rate=1e-5,
- warmup_steps=warmup_steps,
- hold_base_rate_steps=num_train//2,
- min_learn_rate=1e-6
- )
- model.compile(optimizer=Adam(), loss={'yolo_loss': lambda y_true, y_pred: y_pred})
- else:
- print('else')
- reduce_lr = ReduceLROnPlateau(monitor='val_loss', factor=0.5, patience=3, verbose=1)
- model.compile(optimizer=Adam(learning_rate_base), loss={'yolo_loss': lambda y_true, y_pred: y_pred})
-
- print('Train on {} samples, val on {} samples, with batch size {}.'.format(num_train, num_val, batch_size))
- model.fit(data_generator(lines[:num_train], batch_size, input_shape, anchors, num_classes, mosaic=mosaic),
- steps_per_epoch=max(1, num_train//batch_size),
- validation_data=data_generator(lines[num_train:], batch_size, input_shape, anchors, num_classes, mosaic=False),
- validation_steps=max(1, num_val//batch_size),
- epochs=Epoch,
- initial_epoch=Freeze_epoch,
- callbacks=[logging, checkpoint, reduce_lr, early_stopping])
- model.save_weights(log_dir + 'last1.h5')
|