From 6d2d287c9a561ddffd5d2b87c1ff28c3e7fabd17 Mon Sep 17 00:00:00 2001 From: Bubbliiiing <47347516+bubbliiiing@users.noreply.github.com> Date: Mon, 17 May 2021 15:03:20 +0800 Subject: [PATCH] Add files via upload --- FPS_test.py | 1 - get_dr_txt.py | 3 +-- nets/CSPdarknet53.py | 7 +++--- nets/yolo4.py | 6 ++--- train.py | 57 ++++++++++++++++++++++++++------------------ utils/utils.py | 57 ++++++++++++++++++++++++++++++++++++++++++-- voc_annotation.py | 11 +++++---- yolo.py | 12 ++++++---- 8 files changed, 109 insertions(+), 45 deletions(-) diff --git a/FPS_test.py b/FPS_test.py index 1242324..9b4e712 100644 --- a/FPS_test.py +++ b/FPS_test.py @@ -1,4 +1,3 @@ -import os import time import numpy as np diff --git a/get_dr_txt.py b/get_dr_txt.py index dccfda0..8d643f4 100644 --- a/get_dr_txt.py +++ b/get_dr_txt.py @@ -8,8 +8,8 @@ import os import numpy as np from keras import backend as K -from keras.applications.imagenet_utils import preprocess_input from keras.layers import Input +from keras.models import load_model from PIL import Image from tqdm import tqdm @@ -17,7 +17,6 @@ from nets.yolo4 import yolo_body, yolo_eval from utils.utils import letterbox_image from yolo import YOLO - ''' 这里设置的门限值较低是因为计算map需要用到不同门限条件下的Recall和Precision值。 所以只有保留的框足够多，计算的map才会更精确，详情可以了解map的原理。 diff --git a/nets/CSPdarknet53.py b/nets/CSPdarknet53.py index d155fc5..f61b6b2 100644 --- a/nets/CSPdarknet53.py +++ b/nets/CSPdarknet53.py @@ -1,9 +1,8 @@ from functools import wraps from keras import backend as K -from keras.layers import (Add, Concatenate, Conv2D, Layer, MaxPooling2D, - UpSampling2D, ZeroPadding2D) -from keras.layers.advanced_activations import LeakyReLU +from keras.initializers import random_normal +from keras.layers import Add, Concatenate, Conv2D, Layer, ZeroPadding2D from keras.layers.normalization import BatchNormalization from keras.regularizers import l2 from utils.utils import compose @@ -32,7 +31,7 @@ class Mish(Layer): @wraps(Conv2D) def DarknetConv2D(*args, **kwargs): # darknet_conv_kwargs = {'kernel_regularizer': l2(5e-4)} - darknet_conv_kwargs = {} + darknet_conv_kwargs = {'kernel_initializer' : random_normal(stddev=0.02)} darknet_conv_kwargs['padding'] = 'valid' if kwargs.get('strides')==(2,2) else 'same' darknet_conv_kwargs.update(kwargs) return Conv2D(*args, **darknet_conv_kwargs) diff --git a/nets/yolo4.py b/nets/yolo4.py index e62d1e4..cb94e35 100644 --- a/nets/yolo4.py +++ b/nets/yolo4.py @@ -1,10 +1,10 @@ from functools import wraps import keras -import numpy as np import tensorflow as tf from keras import backend as K -from keras.layers import (Add, Concatenate, Conv2D, MaxPooling2D, UpSampling2D, +from keras.initializers import random_normal +from keras.layers import (Concatenate, Conv2D, MaxPooling2D, UpSampling2D, ZeroPadding2D) from keras.layers.advanced_activations import LeakyReLU from keras.layers.normalization import BatchNormalization @@ -23,7 +23,7 @@ from nets.CSPdarknet53 import darknet_body @wraps(Conv2D) def DarknetConv2D(*args, **kwargs): # darknet_conv_kwargs = {'kernel_regularizer': l2(5e-4)} - darknet_conv_kwargs = {} + darknet_conv_kwargs = {'kernel_initializer' : random_normal(stddev=0.02)} darknet_conv_kwargs['padding'] = 'valid' if kwargs.get('strides')==(2,2) else 'same' darknet_conv_kwargs.update(kwargs) return Conv2D(*args, **darknet_conv_kwargs) diff --git a/train.py b/train.py index 573beba..68528af 100644 --- a/train.py +++ b/train.py @@ -1,7 +1,5 @@ import keras.backend as K import numpy as np -import tensorflow as tf -from keras.backend.tensorflow_backend import set_session from keras.callbacks import (EarlyStopping, ModelCheckpoint, ReduceLROnPlateau, TensorBoard) from keras.layers import Input, Lambda @@ -10,8 +8,8 @@ from keras.optimizers import Adam from nets.loss import yolo_loss from nets.yolo4 import yolo_body -from utils.utils import (WarmUpCosineDecayScheduler, get_random_data, - get_random_data_with_Mosaic, rand) +from utils.utils import (LossHistory, WarmUpCosineDecayScheduler, + get_random_data, get_random_data_with_Mosaic) #---------------------------------------------------# @@ -216,7 +214,7 @@ if __name__ == "__main__": # 获取classes和anchor #----------------------------------------------------# class_names = get_classes(classes_path) - anchors = get_anchors(anchors_path) + anchors = get_anchors(anchors_path) #------------------------------------------------------# # 一共有多少类和多少先验框 #------------------------------------------------------# @@ -272,6 +270,7 @@ if __name__ == "__main__": 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) + loss_history = LossHistory(log_dir) #----------------------------------------------------------------------# # 验证集的划分在train.py代码里面进行 @@ -300,20 +299,20 @@ if __name__ == "__main__": # 提示OOM或者显存不足请调小Batch_size #------------------------------------------------------# if True: - Init_epoch = 0 - Freeze_epoch = 50 - batch_size = 8 - learning_rate_base = 1e-3 + Init_epoch = 0 + Freeze_epoch = 50 + batch_size = 8 + learning_rate_base = 1e-3 if Cosine_scheduler: # 预热期 - warmup_epoch = int((Freeze_epoch-Init_epoch)*0.2) + warmup_epoch = int((Freeze_epoch-Init_epoch)*0.2) # 总共的步长 - total_steps = int((Freeze_epoch-Init_epoch) * num_train / batch_size) + total_steps = int((Freeze_epoch-Init_epoch) * num_train / batch_size) # 预热步长 - warmup_steps = int(warmup_epoch * num_train / batch_size) + warmup_steps = int(warmup_epoch * num_train / batch_size) # 学习率 - reduce_lr = WarmUpCosineDecayScheduler(learning_rate_base=learning_rate_base, + reduce_lr = WarmUpCosineDecayScheduler(learning_rate_base=learning_rate_base, total_steps=total_steps, warmup_learning_rate=1e-4, warmup_steps=warmup_steps, @@ -325,23 +324,29 @@ if __name__ == "__main__": 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}) + epoch_size = num_train // batch_size + epoch_size_val = num_val // batch_size + + if epoch_size == 0 or epoch_size_val == 0: + raise ValueError("数据集过小，无法进行训练，请扩充数据集。") + print('Train on {} samples, val on {} samples, with batch size {}.'.format(num_train, num_val, batch_size)) model.fit_generator(data_generator(lines[:num_train], batch_size, input_shape, anchors, num_classes, mosaic=mosaic, random=True), - steps_per_epoch=max(1, num_train//batch_size), + steps_per_epoch=epoch_size, validation_data=data_generator(lines[num_train:], batch_size, input_shape, anchors, num_classes, mosaic=False, random=False), - validation_steps=max(1, num_val//batch_size), + validation_steps=epoch_size_val, epochs=Freeze_epoch, initial_epoch=Init_epoch, - callbacks=[logging, checkpoint, reduce_lr, early_stopping]) + callbacks=[logging, checkpoint, reduce_lr, early_stopping, loss_history]) model.save_weights(log_dir + 'trained_weights_stage_1.h5') for i in range(freeze_layers): model_body.layers[i].trainable = True if True: - Freeze_epoch = 50 - Epoch = 100 - batch_size = 2 - learning_rate_base = 1e-4 + Freeze_epoch = 50 + Epoch = 100 + batch_size = 2 + learning_rate_base = 1e-4 if Cosine_scheduler: # 预热期 @@ -363,12 +368,18 @@ if __name__ == "__main__": 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}) + epoch_size = num_train // batch_size + epoch_size_val = num_val // batch_size + + if epoch_size == 0 or epoch_size_val == 0: + raise ValueError("数据集过小，无法进行训练，请扩充数据集。") + print('Train on {} samples, val on {} samples, with batch size {}.'.format(num_train, num_val, batch_size)) model.fit_generator(data_generator(lines[:num_train], batch_size, input_shape, anchors, num_classes, mosaic=mosaic, random=True), - steps_per_epoch=max(1, num_train//batch_size), + steps_per_epoch=epoch_size, validation_data=data_generator(lines[num_train:], batch_size, input_shape, anchors, num_classes, mosaic=False, random=False), - validation_steps=max(1, num_val//batch_size), + validation_steps=epoch_size_val, epochs=Epoch, initial_epoch=Freeze_epoch, - callbacks=[logging, checkpoint, reduce_lr, early_stopping]) + callbacks=[logging, checkpoint, reduce_lr, early_stopping, loss_history]) model.save_weights(log_dir + 'last1.h5') diff --git a/utils/utils.py b/utils/utils.py index 6be23a8..58d0ce0 100644 --- a/utils/utils.py +++ b/utils/utils.py @@ -1,10 +1,12 @@ +import os from functools import reduce import cv2 import keras import keras.backend as K +import matplotlib.pyplot as plt import numpy as np -from matplotlib.colors import hsv_to_rgb, rgb_to_hsv +import scipy.signal from PIL import Image @@ -298,6 +300,58 @@ def get_random_data(annotation_line, input_shape, max_boxes=100, jitter=.3, hue= return image_data, box_data +class LossHistory(keras.callbacks.Callback): + def __init__(self, log_dir): + import datetime + curr_time = datetime.datetime.now() + time_str = datetime.datetime.strftime(curr_time,'%Y_%m_%d_%H_%M_%S') + self.log_dir = log_dir + self.time_str = time_str + self.save_path = os.path.join(self.log_dir, "loss_" + str(self.time_str)) + self.losses = [] + self.val_loss = [] + + os.makedirs(self.save_path) + + def on_epoch_end(self, batch, logs={}): + self.losses.append(logs.get('loss')) + self.val_loss.append(logs.get('val_loss')) + with open(os.path.join(self.save_path, "epoch_loss_" + str(self.time_str) + ".txt"), 'a') as f: + f.write(str(logs.get('loss'))) + f.write("\n") + with open(os.path.join(self.save_path, "epoch_val_loss_" + str(self.time_str) + ".txt"), 'a') as f: + f.write(str(logs.get('val_loss'))) + f.write("\n") + self.loss_plot() + + def loss_plot(self): + iters = range(len(self.losses)) + + plt.figure() + plt.plot(iters, self.losses, 'red', linewidth = 2, label='train loss') + plt.plot(iters, self.val_loss, 'coral', linewidth = 2, label='val loss') + try: + if len(self.losses) < 25: + num = 5 + else: + num = 15 + + plt.plot(iters, scipy.signal.savgol_filter(self.losses, num, 3), 'green', linestyle = '--', linewidth = 2, label='smooth train loss') + plt.plot(iters, scipy.signal.savgol_filter(self.val_loss, num, 3), '#8B4513', linestyle = '--', linewidth = 2, label='smooth val loss') + except: + pass + + plt.grid(True) + plt.xlabel('Epoch') + plt.ylabel('Loss') + plt.title('A Loss Curve') + plt.legend(loc="upper right") + + plt.savefig(os.path.join(self.save_path, "epoch_loss_" + str(self.time_str) + ".png")) + + plt.cla() + plt.close("all") + def cosine_decay_with_warmup(global_step, learning_rate_base, total_steps, @@ -339,7 +393,6 @@ def cosine_decay_with_warmup(global_step, learning_rate = max(learning_rate,min_learn_rate) return learning_rate - class WarmUpCosineDecayScheduler(keras.callbacks.Callback): """ 继承Callback，实现对学习率的调度 diff --git a/voc_annotation.py b/voc_annotation.py index ca0f88d..324ef31 100644 --- a/voc_annotation.py +++ b/voc_annotation.py @@ -7,7 +7,9 @@ import xml.etree.ElementTree as ET from os import getcwd sets=[('2007', 'train'), ('2007', 'val'), ('2007', 'test')] - +#-----------------------------------------------------# +# 这里设定的classes顺序要和model_data里的txt一样 +#-----------------------------------------------------# classes = ["aeroplane", "bicycle", "bird", "boat", "bottle", "bus", "car", "cat", "chair", "cow", "diningtable", "dog", "horse", "motorbike", "person", "pottedplant", "sheep", "sofa", "train", "tvmonitor"] def convert_annotation(year, image_id, list_file): @@ -19,20 +21,19 @@ def convert_annotation(year, image_id, list_file): difficult = 0 if obj.find('difficult')!=None: difficult = obj.find('difficult').text - cls = obj.find('name').text if cls not in classes or int(difficult)==1: continue cls_id = classes.index(cls) xmlbox = obj.find('bndbox') - b = (int(xmlbox.find('xmin').text), int(xmlbox.find('ymin').text), int(xmlbox.find('xmax').text), int(xmlbox.find('ymax').text)) + b = (int(float(xmlbox.find('xmin').text)), int(float(xmlbox.find('ymin').text)), int(float(xmlbox.find('xmax').text)), int(float(xmlbox.find('ymax').text))) list_file.write(" " + ",".join([str(a) for a in b]) + ',' + str(cls_id)) wd = getcwd() for year, image_set in sets: - image_ids = open('VOCdevkit/VOC%s/ImageSets/Main/%s.txt'%(year, image_set)).read().strip().split() - list_file = open('%s_%s.txt'%(year, image_set), 'w') + image_ids = open('VOCdevkit/VOC%s/ImageSets/Main/%s.txt'%(year, image_set), encoding='utf-8').read().strip().split() + list_file = open('%s_%s.txt'%(year, image_set), 'w', encoding='utf-8') for image_id in image_ids: list_file.write('%s/VOCdevkit/VOC%s/JPEGImages/%s.jpg'%(wd, year, image_id)) convert_annotation(year, image_id, list_file) diff --git a/yolo.py b/yolo.py index f13cddb..164cfd1 100644 --- a/yolo.py +++ b/yolo.py @@ -1,7 +1,5 @@ import colorsys -import copy import os -from timeit import default_timer as timer import numpy as np from keras import backend as K @@ -131,6 +129,11 @@ class YOLO(object): # 检测图片 #---------------------------------------------------# def detect_image(self, image): + #---------------------------------------------------------# + # 在这里将图像转换成RGB图像，防止灰度图在预测时报错。 + #---------------------------------------------------------# + image = image.convert('RGB') + #---------------------------------------------------------# # 给图像增加灰条，实现不失真的resize # 也可以直接resize进行识别 @@ -138,8 +141,7 @@ class YOLO(object): if self.letterbox_image: boxed_image = letterbox_image(image, (self.model_image_size[1],self.model_image_size[0])) else: - boxed_image = image.convert('RGB') - boxed_image = boxed_image.resize((self.model_image_size[1],self.model_image_size[0]), Image.BICUBIC) + boxed_image = image.resize((self.model_image_size[1],self.model_image_size[0]), Image.BICUBIC) image_data = np.array(boxed_image, dtype='float32') image_data /= 255. #---------------------------------------------------------# @@ -162,7 +164,7 @@ class YOLO(object): #---------------------------------------------------------# # 设置字体 #---------------------------------------------------------# - font = ImageFont.truetype(font='font/simhei.ttf', + font = ImageFont.truetype(font='model_data/simhei.ttf', size=np.floor(3e-2 * image.size[1] + 0.5).astype('int32')) thickness = max((image.size[0] + image.size[1]) // 300, 1) -- GitLab