问题遇到的现象和发生背景
在改进一个resnet18的模型的过程中,我们遇到了了一个正确率在80%波动的一个障碍,我们想使得正确率更高,于是想让误差反向传播的过程在一个循环里搞两遍,但是出现了这样的报错,请问如何让误差反向传播过程在一个循环里搞两遍呢?
问题相关代码,请勿粘贴截图
# -*- coding: utf-8 -*-
"""
Created on Fri Jul 22 10:26:33 2022
@author: Blucoris Liang
"""
import torch
import torch.nn as nn
import torch.utils.data as Data
import torchvision.transforms as transforms
import torchvision.datasets as datasets
import torchvision.models as models
EPOCH=30
BATCH_SIZE=40
LR=0.0007
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
train_dataset = datasets.ImageFolder(
'C:\\Users\\19544\\.spyder-py3\\leapGestRecog\\00',
transforms.Compose([
transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
normalize,
]))
train_loader = Data.DataLoader(
train_dataset,
batch_size=BATCH_SIZE,
shuffle=True)
test_loader = Data.DataLoader(
datasets.ImageFolder(
'C:\\Users\\19544\\.spyder-py3\\leapGestRecog\\03',
transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
normalize,
])),
batch_size=BATCH_SIZE, shuffle=False,)
# 数据集长度
train_data_size = len(train_dataset)
print('训练集的长度为:{}'.format(train_data_size))
model = models.resnet18(pretrained=True)
################################
if torch.cuda.is_available(): #
model = models.resnet18(pretrained=True).cuda() #
################################
model.fc = torch.nn.Linear(in_features=512, out_features=10, bias=True).cuda()
fc_params = list(map(id, model.fc.parameters())) # map函数是将fc.parameters()的id返回并组成一个列表
base_params = filter(lambda p: id(p) not in fc_params, model.parameters()) # filter函数是将model.parameters()中地址不在fc.parameters的id中的滤出来
optimizer = torch.optim.SGD([ {'params': base_params}, {'params': model.fc.parameters(), 'lr': LR * 100}], lr=LR)
loss_func=nn.CrossEntropyLoss()
################################
if torch.cuda.is_available(): #
loss_func = loss_func.cuda() #
################################
class AverageMeter(object):
"""Computes and stores the average and current value"""
def __init__(self, name, fmt=':f'):
self.name = name
self.fmt = fmt
self.reset()
def reset(self):
self.val = 0
self.avg = 0
self.sum = 0
self.count = 0
def update(self, val, n=1):
self.val = val
self.sum += val * n
self.count += n
self.avg = self.sum / self.count
def __str__(self):
fmtstr = '{name} {val' + self.fmt + '} ({avg' + self.fmt + '})'
return fmtstr.format(**self.__dict__)
def accuracy(output, target, topk=(1,)):
with torch.no_grad():
maxk = max(topk)
batch_size = target.size(0)
_, pred = output.topk(maxk, 1, True, True)
pred = pred.t()
correct = pred.eq(target.view(1, -1).expand_as(pred))
res = []
for k in topk:
correct_k = correct[:k].view(-1).float().sum(0, keepdim=True)
res.append(correct_k.mul_(100.0 / batch_size))
return res
train_losses = AverageMeter('TrainLoss', ':.4e')
train_top1 = AverageMeter('TrainAccuracy', ':6.2f')
test_losses = AverageMeter('TestLoss', ':.4e')
test_top1 = AverageMeter('TestAccuracy', ':6.2f')
for epoch in range(EPOCH):
model.train()
for i,(images,target) in enumerate(train_loader):
################################
if torch.cuda.is_available(): #
images = images.cuda() #
target = target.cuda() #
################################
output=model(images)
################################
if torch.cuda.is_available(): #
output = output.cuda() #
################################
loss= loss_func(output,target)
acc1, = accuracy(output, target, topk=(1,))
train_losses.update(loss.item(), images.size(0))
train_top1.update(acc1[0], images.size(0))
# 反向传播第一遍
optimizer.zero_grad()
loss.backward(retain_graph = True)
optimizer.step()
torch.autograd.set_detect_anomaly(True)
# 反向传播第二遍
optimizer.zero_grad()
loss.backward()
optimizer.step()
print('Epoch[{}/{}],TrainLoss:{}, TrainAccuracy:{}'.format(epoch,EPOCH,train_losses.val, train_top1.val))
model.eval()
with torch.no_grad():
for i,(images,target) in enumerate(test_loader):
################################
if torch.cuda.is_available(): #
images = images.cuda() #
target = target.cuda() #
################################
output=model(images)
loss= loss_func(output,target)
acc1, = accuracy(output, target, topk=(1,))
test_losses.update(loss.item(), images.size(0))
test_top1.update(acc1[0], images.size(0))
print('TestLoss:{}, TestAccuracy:{}'.format(test_losses.avg, test_top1.avg))
运行结果及报错内容
runfile('C:/Users/19544/.spyder-py3/成功对手势识别用resnet进行了第一次训练.py', wdir='C:/Users/19544/.spyder-py3')
训练集的长度为:2000
D:\ANACONDA\envs\MyEnv\lib\site-packages\torch\autograd\__init__.py:173: UserWarning: Error detected in AddmmBackward0. No forward pass information available. Enable detect anomaly during forward pass for more information. (Triggered internally at C:\actions-runner\_work\pytorch\pytorch\builder\windows\pytorch\torch\csrc\autograd\python_anomaly_mode.cpp:85.)
Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
Traceback (most recent call last):
File "C:\Users\19544\.spyder-py3\成功对手势识别用resnet进行了第一次训练.py", line 146, in <module>
loss.backward()
File "D:\ANACONDA\envs\MyEnv\lib\site-packages\torch\_tensor.py", line 363, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs)
File "D:\ANACONDA\envs\MyEnv\lib\site-packages\torch\autograd\__init__.py", line 173, in backward
Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [512, 10]], which is output 0 of AsStridedBackward0, is at version 2; expected version 1 instead. Hint: the backtrace further above shows the operation that failed to compute its gradient. The variable in question was changed in there or anywhere later. Good luck!
我的解答思路和尝试过的方法
之前尝试过添加上了retain_graph = True,好像也不行
我想要达到的结果
成功实现在一个循环里误差反向传播两次,或者希望能获取大家的其他提高正确率的好方法