语义分割训练模型代码为下篇学习解读做准备

这是我下一篇要学习解读的训练代码&＃xff1a;

from __future__ import print_function
import os,time,cv2, sys, math
import tensorflow as tf
import tensorflow.contrib.slim as slim
import numpy as np
import time, datetime
import argparse
import random
import os, sys
import subprocess# use &＃39;Agg&＃39; on matplotlib so that plots could be generated even without Xserver
# running
import matplotlib
matplotlib.use(&＃39;Agg&＃39;)from utils import utils, helpers
from builders import model_builderimport matplotlib.pyplot as pltdef str2bool(v):if v.lower() in (&＃39;yes&＃39;, &＃39;true&＃39;, &＃39;t&＃39;, &＃39;y&＃39;, &＃39;1&＃39;):return Trueelif v.lower() in (&＃39;no&＃39;, &＃39;false&＃39;, &＃39;f&＃39;, &＃39;n&＃39;, &＃39;0&＃39;):return Falseelse:raise argparse.ArgumentTypeError(&＃39;Boolean value expected.&＃39;)parser &＃61; argparse.ArgumentParser()
parser.add_argument(&＃39;--num_epochs&＃39;, type&＃61;int, default&＃61;30, help&＃61;&＃39;Number of epochs to train for&＃39;)
parser.add_argument(&＃39;--epoch_start_i&＃39;, type&＃61;int, default&＃61;0, help&＃61;&＃39;Start counting epochs from this number&＃39;)
parser.add_argument(&＃39;--checkpoint_step&＃39;, type&＃61;int, default&＃61;5, help&＃61;&＃39;How often to save checkpoints (epochs)&＃39;)
parser.add_argument(&＃39;--validation_step&＃39;, type&＃61;int, default&＃61;1, help&＃61;&＃39;How often to perform validation (epochs)&＃39;)
parser.add_argument(&＃39;--image&＃39;, type&＃61;str, default&＃61;None, help&＃61;&＃39;The image you want to predict on. Only valid in "predict" mode.&＃39;)
parser.add_argument(&＃39;--continue_training&＃39;, type&＃61;str2bool, default&＃61;False, help&＃61;&＃39;Whether to continue training from a checkpoint&＃39;)
parser.add_argument(&＃39;--dataset&＃39;, type&＃61;str, default&＃61;"CamVid", help&＃61;&＃39;Dataset you are using.&＃39;)
parser.add_argument(&＃39;--crop_height&＃39;, type&＃61;int, default&＃61;512, help&＃61;&＃39;Height of cropped input image to network&＃39;)
parser.add_argument(&＃39;--crop_width&＃39;, type&＃61;int, default&＃61;512, help&＃61;&＃39;Width of cropped input image to network&＃39;)
parser.add_argument(&＃39;--batch_size&＃39;, type&＃61;int, default&＃61;1, help&＃61;&＃39;Number of images in each batch&＃39;)
parser.add_argument(&＃39;--num_val_images&＃39;, type&＃61;int, default&＃61;20, help&＃61;&＃39;The number of images to used for validations&＃39;)
parser.add_argument(&＃39;--h_flip&＃39;, type&＃61;str2bool, default&＃61;False, help&＃61;&＃39;Whether to randomly flip the image horizontally for data augmentation&＃39;)
parser.add_argument(&＃39;--v_flip&＃39;, type&＃61;str2bool, default&＃61;False, help&＃61;&＃39;Whether to randomly flip the image vertically for data augmentation&＃39;)
parser.add_argument(&＃39;--brightness&＃39;, type&＃61;float, default&＃61;None, help&＃61;&＃39;Whether to randomly change the image brightness for data augmentation. Specifies the max bightness change as a factor between 0.0 and 1.0. For example, 0.1 represents a max brightness change of 10%% (&＃43;-).&＃39;)
parser.add_argument(&＃39;--rotation&＃39;, type&＃61;float, default&＃61;None, help&＃61;&＃39;Whether to randomly rotate the image for data augmentation. Specifies the max rotation angle in degrees.&＃39;)
parser.add_argument(&＃39;--model&＃39;, type&＃61;str, default&＃61;"DeepLabV3", help&＃61;&＃39;The model you are using. See model_builder.py for supported models&＃39;)
parser.add_argument(&＃39;--frontend&＃39;, type&＃61;str, default&＃61;"ResNet101", help&＃61;&＃39;The frontend you are using. See frontend_builder.py for supported models&＃39;)
args &＃61; parser.parse_args()def data_augmentation(input_image, output_image):# Data augmentationinput_image, output_image &＃61; utils.random_crop(input_image, output_image, args.crop_height, args.crop_width)if args.h_flip and random.randint(0,1):input_image &＃61; cv2.flip(input_image, 1)output_image &＃61; cv2.flip(output_image, 1)if args.v_flip and random.randint(0,1):input_image &＃61; cv2.flip(input_image, 0)output_image &＃61; cv2.flip(output_image, 0)if args.brightness:factor &＃61; 1.0 &＃43; random.uniform(-1.0*args.brightness, args.brightness)table &＃61; np.array([((i / 255.0) * factor) * 255 for i in np.arange(0, 256)]).astype(np.uint8)input_image &＃61; cv2.LUT(input_image, table)if args.rotation:angle &＃61; random.uniform(-1*args.rotation, args.rotation)if args.rotation:M &＃61; cv2.getRotationMatrix2D((input_image.shape[1]//2, input_image.shape[0]//2), angle, 1.0)input_image &＃61; cv2.warpAffine(input_image, M, (input_image.shape[1], input_image.shape[0]), flags&＃61;cv2.INTER_NEAREST)output_image &＃61; cv2.warpAffine(output_image, M, (output_image.shape[1], output_image.shape[0]), flags&＃61;cv2.INTER_NEAREST)return input_image, output_image# Get the names of the classes so we can record the evaluation results
class_names_list, label_values &＃61; helpers.get_label_info(os.path.join(args.dataset, "class_dict.csv"))
class_names_string &＃61; ""
for class_name in class_names_list:if not class_name &＃61;&＃61; class_names_list[-1]:class_names_string &＃61; class_names_string &＃43; class_name &＃43; ", "else:class_names_string &＃61; class_names_string &＃43; class_namenum_classes &＃61; len(label_values)config &＃61; tf.ConfigProto()
config.gpu_options.allow_growth &＃61; True
sess&＃61;tf.Session(config&＃61;config)# Compute your softmax cross entropy loss
net_input &＃61; tf.placeholder(tf.float32,shape&＃61;[None,None,None,3])
net_output &＃61; tf.placeholder(tf.float32,shape&＃61;[None,None,None,num_classes])network, init_fn &＃61; model_builder.build_model(model_name&＃61;args.model, frontend&＃61;args.frontend, net_input&＃61;net_input, num_classes&＃61;num_classes, crop_width&＃61;args.crop_width, crop_height&＃61;args.crop_height, is_training&＃61;True)loss &＃61; tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits&＃61;network, labels&＃61;net_output))opt &＃61; tf.train.RMSPropOptimizer(learning_rate&＃61;0.0001, decay&＃61;0.995).minimize(loss, var_list&＃61;[var for var in tf.trainable_variables()])saver&＃61;tf.train.Saver(max_to_keep&＃61;1000)
sess.run(tf.global_variables_initializer())utils.count_params()# If a pre-trained ResNet is required, load the weights.
# This must be done AFTER the variables are initialized with sess.run(tf.global_variables_initializer())
if init_fn is not None:init_fn(sess)# Load a previous checkpoint if desired
model_checkpoint_name &＃61; "checkpoints/latest_model_" &＃43; args.model &＃43; "_" &＃43; args.dataset &＃43; ".ckpt"
if args.continue_training:print(&＃39;Loaded latest model checkpoint&＃39;)saver.restore(sess, model_checkpoint_name)# Load the data
print("Loading the data ...")
train_input_names,train_output_names, val_input_names, val_output_names, test_input_names, test_output_names &＃61; utils.prepare_data(dataset_dir&＃61;args.dataset)print("\n***** Begin training *****")
print("Dataset -->", args.dataset)
print("Model -->", args.model)
print("Crop Height -->", args.crop_height)
print("Crop Width -->", args.crop_width)
print("Num Epochs -->", args.num_epochs)
print("Batch Size -->", args.batch_size)
print("Num Classes -->", num_classes)print("Data Augmentation:")
print("\tVertical Flip -->", args.v_flip)
print("\tHorizontal Flip -->", args.h_flip)
print("\tBrightness Alteration -->", args.brightness)
print("\tRotation -->", args.rotation)
print("")avg_loss_per_epoch &＃61; []
avg_scores_per_epoch &＃61; []
avg_iou_per_epoch &＃61; []# Which validation images do we want
val_indices &＃61; []
num_vals &＃61; min(args.num_val_images, len(val_input_names))# Set random seed to make sure models are validated on the same validation images.
# So you can compare the results of different models more intuitively.
random.seed(16)
val_indices&＃61;random.sample(range(0,len(val_input_names)),num_vals)# Do the training here
Animal &＃61; []
Archway &＃61; []
Bicyclist &＃61; []
Bridge &＃61; []
Building &＃61; []
Car &＃61; []
CartLuggagePram &＃61; []
Child &＃61; []
Column_Pole &＃61; []
Fence &＃61; []
LaneMkgsDriv &＃61; []
LaneMkgsNonDriv &＃61; []
Misc_Text &＃61; []
MotorcycleScooter &＃61; []
OtherMoving &＃61; []
ParkingBlock &＃61; []
Pedestrian &＃61; []
Road &＃61; []
RoadShoulder &＃61; []
Sidewalk &＃61; []
SignSymbol &＃61; []
Sky &＃61; []
SUVPickupTruck &＃61; []
TrafficCone &＃61; []
TrafficLight &＃61; []
Train &＃61; []
Tree &＃61; []
Truck_Bus &＃61; []
Tunnel &＃61; []
VegetationMisc &＃61; []
Void &＃61; []
Wall &＃61; []for epoch in range(args.epoch_start_i, args.num_epochs):current_losses &＃61; []cnt&＃61;0# Equivalent to shufflingid_list &＃61; np.random.permutation(len(train_input_names))num_iters &＃61; int(np.floor(len(id_list) / args.batch_size))st &＃61; time.time()epoch_st&＃61;time.time()for i in range(num_iters):# st&＃61;time.time()input_image_batch &＃61; []output_image_batch &＃61; []# Collect a batch of imagesfor j in range(args.batch_size):index &＃61; i*args.batch_size &＃43; jid &＃61; id_list[index]input_image &＃61; utils.load_image(train_input_names[id])output_image &＃61; utils.load_image(train_output_names[id])with tf.device(&＃39;/cpu:0&＃39;):input_image, output_image &＃61; data_augmentation(input_image, output_image)# Prep the data. Make sure the labels are in one-hot formatinput_image &＃61; np.float32(input_image) / 255.0output_image &＃61; np.float32(helpers.one_hot_it(label&＃61;output_image, label_values&＃61;label_values))input_image_batch.append(np.expand_dims(input_image, axis&＃61;0))output_image_batch.append(np.expand_dims(output_image, axis&＃61;0))if args.batch_size &＃61;&＃61; 1:input_image_batch &＃61; input_image_batch[0]output_image_batch &＃61; output_image_batch[0]else:input_image_batch &＃61; np.squeeze(np.stack(input_image_batch, axis&＃61;1))output_image_batch &＃61; np.squeeze(np.stack(output_image_batch, axis&＃61;1))# Do the training_,current&＃61;sess.run([opt,loss],feed_dict&＃61;{net_input:input_image_batch,net_output:output_image_batch})current_losses.append(current)cnt &＃61; cnt &＃43; args.batch_sizeif cnt % 20 &＃61;&＃61; 0:string_print &＃61; "Epoch &＃61; %d Count &＃61; %d Current_Loss &＃61; %.4f Time &＃61; %.2f"%(epoch,cnt,current,time.time()-st)utils.LOG(string_print)st &＃61; time.time()mean_loss &＃61; np.mean(current_losses)avg_loss_per_epoch.append(mean_loss)# Create directories if neededif not os.path.isdir("%s/%04d"%("checkpoints",epoch)):os.makedirs("%s/%04d"%("checkpoints",epoch))# Save latest checkpoint to same file nameprint("Saving latest checkpoint")saver.save(sess,model_checkpoint_name)if val_indices !&＃61; 0 and epoch % args.checkpoint_step &＃61;&＃61; 0:print("Saving checkpoint for this epoch")saver.save(sess,"%s/%04d/model.ckpt"%("checkpoints",epoch))if epoch % args.validation_step &＃61;&＃61; 0:print("Performing validation")target&＃61;open("%s/%04d/val_scores.csv"%("checkpoints",epoch),&＃39;w&＃39;)target.write("val_name, avg_accuracy, precision, recall, f1 score, mean iou, %s\n" % (class_names_string))scores_list &＃61; []class_scores_list &＃61; []precision_list &＃61; []recall_list &＃61; []f1_list &＃61; []iou_list &＃61; []# Do the validation on a small set of validation imagesfor ind in val_indices:input_image &＃61; np.expand_dims(np.float32(utils.load_image(val_input_names[ind])[:args.crop_height, :args.crop_width]),axis&＃61;0)/255.0gt &＃61; utils.load_image(val_output_names[ind])[:args.crop_height, :args.crop_width]gt &＃61; helpers.reverse_one_hot(helpers.one_hot_it(gt, label_values))# st &＃61; time.time()output_image &＃61; sess.run(network,feed_dict&＃61;{net_input:input_image})output_image &＃61; np.array(output_image[0,:,:,:])output_image &＃61; helpers.reverse_one_hot(output_image)out_vis_image &＃61; helpers.colour_code_segmentation(output_image, label_values)accuracy, class_accuracies, prec, rec, f1, iou &＃61; utils.evaluate_segmentation(pred&＃61;output_image, label&＃61;gt, num_classes&＃61;num_classes)file_name &＃61; utils.filepath_to_name(val_input_names[ind])target.write("%s, %f, %f, %f, %f, %f"%(file_name, accuracy, prec, rec, f1, iou))for item in class_accuracies:target.write(", %f"%(item))target.write("\n")scores_list.append(accuracy)class_scores_list.append(class_accuracies)precision_list.append(prec)recall_list.append(rec)f1_list.append(f1)iou_list.append(iou)gt &＃61; helpers.colour_code_segmentation(gt, label_values)file_name &＃61; os.path.basename(val_input_names[ind])file_name &＃61; os.path.splitext(file_name)[0]cv2.imwrite("%s/%04d/%s_pred.png"%("checkpoints",epoch, file_name),cv2.cvtColor(np.uint8(out_vis_image), cv2.COLOR_RGB2BGR))cv2.imwrite("%s/%04d/%s_gt.png"%("checkpoints",epoch, file_name),cv2.cvtColor(np.uint8(gt), cv2.COLOR_RGB2BGR))target.close()avg_score &＃61; np.mean(scores_list)class_avg_scores &＃61; np.mean(class_scores_list, axis&＃61;0)avg_scores_per_epoch.append(avg_score)avg_precision &＃61; np.mean(precision_list)avg_recall &＃61; np.mean(recall_list)avg_f1 &＃61; np.mean(f1_list)avg_iou &＃61; np.mean(iou_list)avg_iou_per_epoch.append(avg_iou)print("\nAverage validation accuracy for epoch # %04d &＃61; %f"% (epoch, avg_score))print("Average per class validation accuracies for epoch # %04d:"% (epoch))for index, item in enumerate(class_avg_scores):print("%s &＃61; %f" % (class_names_list[index], item))print("Validation precision &＃61; ", avg_precision)print("Validation recall &＃61; ", avg_recall)print("Validation F1 score &＃61; ", avg_f1)print("Validation IoU score &＃61; ", avg_iou)epoch_time&＃61;time.time()-epoch_stremain_time&＃61;epoch_time*(args.num_epochs-1-epoch)m, s &＃61; divmod(remain_time, 60)h, m &＃61; divmod(m, 60)if s!&＃61;0:train_time&＃61;"Remaining training time &＃61; %d hours %d minutes %d seconds\n"%(h,m,s)else:train_time&＃61;"Remaining training time : Training completed.\n"utils.LOG(train_time)scores_list &＃61; []fig1, ax1 &＃61; plt.subplots(figsize&＃61;(11, 8))ax1.plot(range(epoch&＃43;1), avg_scores_per_epoch)ax1.set_title("Average validation accuracy vs epochs")ax1.set_xlabel("Epoch")ax1.set_ylabel("Avg. val. accuracy")plt.savefig(&＃39;accuracy_vs_epochs.png&＃39;)plt.clf()fig2, ax2 &＃61; plt.subplots(figsize&＃61;(11, 8))ax2.plot(range(epoch&＃43;1), avg_loss_per_epoch)ax2.set_title("Average loss vs epochs")ax2.set_xlabel("Epoch")ax2.set_ylabel("Current loss")plt.savefig(&＃39;loss_vs_epochs.png&＃39;)plt.clf()fig3, ax3 &＃61; plt.subplots(figsize&＃61;(11, 8))ax3.plot(range(epoch&＃43;1), avg_iou_per_epoch)ax3.set_title("Average IoU vs epochs")ax3.set_xlabel("Epoch")ax3.set_ylabel("Current IoU")plt.savefig(&＃39;iou_vs_epochs.png&＃39;)for index, item in enumerate(class_avg_scores):print(item)print(class_names_list[index])if "Animal" &＃61;&＃61; class_names_list[index]:Animal.append(item)print(Animal)print(class_names_list[index])fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Animal)elif "Archway" &＃61;&＃61; class_names_list[index]:Archway.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Archway)elif "Bicyclist" &＃61;&＃61; class_names_list[index]:Bicyclist.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Bicyclist)elif "Bridge" &＃61;&＃61; class_names_list[index]:Bridge.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Bridge)elif "Building" &＃61;&＃61; class_names_list[index]:Building.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Building)elif "Car" &＃61;&＃61; class_names_list[index]:Car .append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Car )elif "CartLuggagePram" &＃61;&＃61; class_names_list[index]:CartLuggagePram.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), CartLuggagePram)elif "Child" &＃61;&＃61; class_names_list[index]:Child.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Child)elif "Column_Pole" &＃61;&＃61; class_names_list[index]:Column_Pole.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Column_Pole)elif "Fence" &＃61;&＃61; class_names_list[index]:Fence.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Fence)elif "LaneMkgsDriv" &＃61;&＃61; class_names_list[index]:LaneMkgsDriv.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), LaneMkgsDriv)elif "LaneMkgsNonDriv" &＃61;&＃61; class_names_list[index]:LaneMkgsNonDriv.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), LaneMkgsNonDriv)elif "Misc_Text" &＃61;&＃61; class_names_list[index]:Misc_Text.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Misc_Text)elif "MotorcycleScooter" &＃61;&＃61; class_names_list[index]:MotorcycleScooter.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), MotorcycleScooter)elif "OtherMoving" &＃61;&＃61; class_names_list[index]:OtherMoving.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), OtherMoving)elif "ParkingBlock" &＃61;&＃61; class_names_list[index]:ParkingBlock.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), ParkingBlock)elif "Pedestrian" &＃61;&＃61; class_names_list[index]:Pedestrian.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Pedestrian)elif "Road" &＃61;&＃61; class_names_list[index]:Road.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Road)elif "RoadShoulder" &＃61;&＃61; class_names_list[index]:RoadShoulder.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), RoadShoulder)elif "Sidewalk" &＃61;&＃61; class_names_list[index]:Sidewalk.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Sidewalk)elif "SignSymbol" &＃61;&＃61; class_names_list[index]:SignSymbol.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), SignSymbol)elif "Sky" &＃61;&＃61; class_names_list[index]:Sky.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Sky)elif "SUVPickupTruck" &＃61;&＃61; class_names_list[index]:SUVPickupTruck.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), SUVPickupTruck)elif "TrafficCone" &＃61;&＃61; class_names_list[index]:TrafficCone .append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), TrafficCone )elif "TrafficLight" &＃61;&＃61; class_names_list[index]:TrafficLight .append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), TrafficLight )elif "Train" &＃61;&＃61; class_names_list[index]:Train.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Train)elif "Tree" &＃61;&＃61; class_names_list[index]:Tree.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Tree)elif "Truck_Bus" &＃61;&＃61; class_names_list[index]:Truck_Bus.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Truck_Bus)elif "Tunnel" &＃61;&＃61; class_names_list[index]:Tunnel.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Tunnel)elif "VegetationMisc" &＃61;&＃61; class_names_list[index]:VegetationMisc.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), VegetationMisc)elif "Void" &＃61;&＃61; class_names_list[index]:Void.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Void)elif "Wall" &＃61;&＃61; class_names_list[index]:Wall.append(item)fig, ax &＃61; plt.subplots(figsize&＃61;(11, 8))ax.plot(range(epoch &＃43; 1), Wall)else:print("fuck")ax.set_title("%s vs epochs"% class_names_list[index])ax.set_xlabel("Epoch")ax.set_ylabel("accuracy")plt.savefig("%s_accuracy_vs_epochs.png"% class_names_list[index])

参考&＃xff1a;https://blog.csdn.net/qq_34606546/article/details/86514306