模型做预测时,各种值的打印。
MobileNetV2模型预测单张图:
import os
import json
import torch
from PIL import Image
from torchvision import transforms
import matplotlib.pyplot as plt
from model_v2 import MobileNetV2
def main():
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
data_transform = transforms.Compose(
[transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])
# load image
img_path = "2.jpg"
assert os.path.exists(img_path), "file: '{}' dose not exist.".format(img_path)
img = Image.open(img_path)
plt.imshow(img)
# [N, C, H, W]
img = data_transform(img)
# expand batch dimension
img = torch.unsqueeze(img, dim=0)
print("img:",img) # [batchsize,c,h,w]
# read class_indict
json_path = './class_indices.json'
assert os.path.exists(json_path), "file: '{}' dose not exist.".format(json_path)
with open(json_path, "r") as f:
class_indict = json.load(f)
# create model
model = MobileNetV2(num_classes=5).to(device)
# load model weights
model_weight_path = "./MobileNetV2.pth"
model.load_state_dict(torch.load(model_weight_path, map_location=device))
model.eval()
with torch.no_grad():
# predict class
output = model(img.to(device)).cpu()
print("output:",output)
print("output.shape:",output.shape)
output = torch.squeeze(output) #压缩batch维度
predict = torch.softmax(output, dim=0)#从scores变成概率
predict_cla = torch.argmax(predict).numpy()
print_res = "class: {} prob: {:.3}".format(class_indict[str(predict_cla)],
predict[predict_cla].numpy())
plt.title(print_res)
print(predict)
for i in range(len(predict)):
print("class: {:10} prob: {:.3}".format(class_indict[str(i)],
predict[i].numpy()))
plt.show()
if __name__ == '__main__':
main()
输出:
img: tensor([[[[ 1.9407, 1.9407, 1.9749, ..., -0.0458, 0.9303, 1.4612], #增加了batch维度,目前是只有一张图,
[ 1.8893, 1.9407, 1.9578, ..., -0.5082, -0.0287, 0.8789], # 但放到模型里需要batch维度
[ 1.8379, 1.8722, 1.8037, ..., -0.1828, -0.3883, -0.0458],
...,
[-1.7069, -1.6727, -1.6898, ..., -0.5082, 0.2967, -0.0287],
[-1.7412, -1.7069, -1.7412, ..., -0.4911, 0.4679, -0.4739],
[-1.7412, -1.7925, -1.7925, ..., -0.3712, 0.1254, -0.1143]],
[[ 2.4111, 2.3585, 2.2710, ..., -0.7227, 0.2577, 1.2906],
[ 2.2710, 2.2360, 2.2010, ..., -0.7402, -0.8277, 0.0476],
[ 2.0259, 2.0084, 1.9734, ..., -0.2850, -0.6702, -0.7052],
...,
[-1.6155, -1.7031, -1.6155, ..., -0.3375, 0.3452, 0.1352],
[-1.6856, -1.6681, -1.7031, ..., -0.3200, 0.3627, 0.2052],
[-1.7381, -1.7206, -1.7206, ..., -0.0574, 0.2402, 0.1527]],
[[-0.4450, -0.0790, -0.0615, ..., -1.1073, -0.7238, 0.6879],
[-0.5321, -0.1312, -0.2010, ..., -0.7587, -1.2467, -0.7936],
[-0.3055, -0.0790, -0.2010, ..., -0.0092, -0.4973, -1.0376],
...,
[-1.3513, -1.4559, -1.4384, ..., -0.0790, 0.5136, 0.4439],
[-1.4733, -1.4733, -1.4907, ..., -0.0790, 0.5311, 0.5136],
[-1.4210, -1.4036, -1.4907, ..., 0.3393, 0.4265, 0.3219]]]])
output: tensor([[ 0.4225, -1.3952, -1.5308, -0.4893, -2.1681]])#真正的
output.shape: torch.Size([1, 5])
tensor([0.5615, 0.0912, 0.0796, 0.2256, 0.0421]) #先UNsqueeze了batch维度,再经过softmax变成 预测概率
class: daisy prob: 0.561
class: dandelion prob: 0.0912
class: roses prob: 0.0796
class: sunflowers prob: 0.226
class: tulips prob: 0.0421
多张图预测见下篇博客数据集导入预处理和多张图片预测
使用CIFAR10做样本:
transform = transforms.Compose(
[transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
# 50000张训练图片
# 第一次使用时要将download设置为True才会自动去下载数据集
#图片和对应label值都在这得到了
train_set = torchvision.datasets.CIFAR10(root='./data', train=True,
download=False, transform=transform)
train_loader = torch.utils.data.DataLoader(train_set, batch_size=36,#一个batch里的图片数
shuffle=True, num_workers=0)
# 10000张验证图片
# 第一次使用时要将download设置为True才会自动去下载数据集
val_set = torchvision.datasets.CIFAR10(root='./data', train=False,
download=False, transform=transform)
val_loader = torch.utils.data.DataLoader(val_set, batch_size=10000,
shuffle=False, num_workers=0)#train_dataset随机翻转和裁剪是为了抑制过拟合,提升模型泛化能力。val_dataset当中无此操作,只是resize到224作为模型的输入尺寸
val_data_iter = iter(val_loader)#转化为迭代器
val_image, val_label = val_data_iter.next()#得到一批图像和标签
加载自己的样本做多batch的训练时
data_transform = {
"train": transforms.Compose([transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])]),
"val": transforms.Compose([transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])}
data_root = os.path.abspath(os.path.join(os.getcwd(), "../..")) # get data root path
image_path = os.path.join(data_root, "data_set", "flower_data") # flower data set path
assert os.path.exists(image_path), "{} path does not exist.".format(image_path)
train_dataset = datasets.ImageFolder(root=os.path.join(image_path, "train"),#加载train数据集
transform=data_transform["train"])#处理方式
train_num = len(train_dataset)
# {'daisy':0, 'dandelion':1, 'roses':2, 'sunflower':3, 'tulips':4}
flower_list = train_dataset.class_to_idx # 分类名称对应的索引
cla_dict = dict((val, key) for key, val in flower_list.items()) # 将key,value值反过来,已达到通过索引找到分类的目的
# write dict into json file
json_str = json.dumps(cla_dict, indent=4) # 编码成json格式
with open('class_indices.json', 'w') as json_file: # 写进去
json_file.write(json_str)
#nw = min([os.cpu_count(), batch_size if batch_size > 1 else 0, 8]) # number of workers
nw = 0
print('Using {} dataloader workers every process'.format(nw))
train_loader = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size, shuffle=True,
num_workers=nw)
validate_dataset = datasets.ImageFolder(root=os.path.join(image_path, "val"),
transform=data_transform["val"])
val_num = len(validate_dataset)
validate_loader = torch.utils.data.DataLoader(validate_dataset,
batch_size=batch_size, shuffle=False,
num_workers=nw)
print("using {} images for training, {} images for validation.".format(train_num,
val_num))
训练和验证样本的调用
(好几个batch的图片都放在train_loader和val_loader里,使用时用
for val_data in val_loader或for step, data in enumerate(train_loader)来循环得到每个batch里的图片)
train_steps = len(train_loader)
for epoch in range(epochs):
# train
net.train()
running_loss = 0.0
train_bar = tqdm(train_loader, file=sys.stdout) #加载条,可以去掉,更改成
for step, data in enumerate(train_bar): #enumerate(train_loader)
images, labels = data
optimizer.zero_grad()
logits = net(images.to(device))
loss = loss_function(logits, labels.to(device))
loss.backward()
optimizer.step()
# print statistics
running_loss += loss.item()
train_bar.desc = "train epoch[{}/{}] loss:{:.3f}".format(epoch + 1,
epochs,
loss)
# validate
net.eval()
acc = 0.0 # accumulate accurate number / epoch
with torch.no_grad():
val_bar = tqdm(validate_loader, file=sys.stdout)
for val_data in val_bar:
val_images, val_labels = val_data
outputs = net(val_images.to(device))
# loss = loss_function(outputs, test_labels)
predict_y = torch.max(outputs, dim=1)[1]
acc += torch.eq(predict_y, val_labels.to(device)).sum().item()
val_bar.desc = "valid epoch[{}/{}]".format(epoch + 1,
epochs)
val_accurate = acc / val_num
print('[epoch %d] train_loss: %.3f val_accuracy: %.3f' %
(epoch + 1, running_loss / train_steps, val_accurate))
if val_accurate > best_acc:#在正确率最高的时候保存
best_acc = val_accurate
torch.save(net.state_dict(), save_path)
print('Finished Training')