快捷導(dǎo)航

pytorch實(shí)現(xiàn)對(duì)輸入超過三通道的數(shù)據(jù)進(jìn)行訓(xùn)練

更新時(shí)間：2020年01月15日 10:06:51 作者：東城青年

今天小編就為大家分享一篇pytorch實(shí)現(xiàn)對(duì)輸入超過三通道的數(shù)據(jù)進(jìn)行訓(xùn)練，具有很好的參考價(jià)值，希望對(duì)大家有所幫助。一起跟隨小編過來看看吧

案例背景:視頻識(shí)別

假設(shè)每次輸入是8s的灰度視頻,視頻幀率為25fps,則視頻由200幀圖像序列構(gòu)成.每幀是一副單通道的灰度圖像,通過pythonb里面的np.stack(深度拼接)可將200幀拼接成200通道的深度數(shù)據(jù).進(jìn)而送到網(wǎng)絡(luò)里面去訓(xùn)練.

如果輸入圖像200通道覺得多,可以對(duì)視頻進(jìn)行抽幀,針對(duì)具體場(chǎng)景可以隨機(jī)抽幀或等間隔抽幀.比如這里等間隔抽取40幀.則最后輸入視頻相當(dāng)于輸入一個(gè)40通道的圖像數(shù)據(jù)了.

pytorch對(duì)超過三通道數(shù)據(jù)的加載:

讀取視頻每一幀,轉(zhuǎn)為array格式,然后依次將每一幀進(jìn)行深度拼接,最后得到一個(gè)40通道的array格式的深度數(shù)據(jù),保存到pickle里.

對(duì)每個(gè)視頻都進(jìn)行上述操作,保存到pickle里.

我這里將火的視頻深度數(shù)據(jù)保存在一個(gè).pkl文件中,一共2504個(gè)火的視頻,即2504個(gè)火的深度數(shù)據(jù).

將非火的視頻深度數(shù)據(jù)保存在一個(gè).pkl文件中,一共3985個(gè)非火的視頻,即3985個(gè)非火的深度數(shù)據(jù).

數(shù)據(jù)加載

import torch 
from torch.utils import data
import os
from PIL import Image
import numpy as np
import pickle
 
class Fire_Unfire(data.Dataset):
  def __init__(self,fire_path,unfire_path):
    self.pickle_fire = open(fire_path,'rb')
    self.pickle_unfire = open(unfire_path,'rb')
    
  def __getitem__(self,index):
    if index <2504:
      fire = pickle.load(self.pickle_fire)#高*寬*通道
      fire = fire.transpose(2,0,1)#通道*高*寬
      data = torch.from_numpy(fire)
      label = 1
      return data,label
    elif index>=2504 and index<6489:
      unfire = pickle.load(self.pickle_unfire)
      unfire = unfire.transpose(2,0,1)
      data = torch.from_numpy(unfire)
      label = 0
      return data,label
    
  def __len__(self):
    return 6489

root_path = './datasets/train'
dataset = Fire_Unfire(root_path +'/fire_train.pkl',root_path +'/unfire_train.pkl')
 
#轉(zhuǎn)換成pytorch網(wǎng)絡(luò)輸入的格式(批量大小,通道數(shù),高,寬)
from torch.utils.data import DataLoader
fire_dataloader = DataLoader(dataset,batch_size=4,shuffle=True,drop_last = True)

模型訓(xùn)練

import torch
from torch.utils import data
from nets.mobilenet import mobilenet
from config.config import default_config
from torch.autograd import Variable as V
import numpy as np
import sys
import time
 
opt = default_config()
def train():
  #模型定義
  model = mobilenet().cuda()
  if opt.pretrain_model:
    model.load_state_dict(torch.load(opt.pretrain_model))
  
  #損失函數(shù)
  criterion = torch.nn.CrossEntropyLoss().cuda()
  
  #學(xué)習(xí)率
  lr = opt.lr
  
  #優(yōu)化器
  optimizer = torch.optim.SGD(model.parameters(),lr = lr,weight_decay=opt.weight_decay)
  
  
  pre_loss = 0.0
  #訓(xùn)練
  for epoch in range(opt.max_epoch):
     #訓(xùn)練數(shù)據(jù)
    train_data = Fire_Unfire(opt.root_path +'/fire_train.pkl',opt.root_path +'/unfire_train.pkl')
    train_dataloader = data.DataLoader(train_data,batch_size=opt.batch_size,shuffle=True,drop_last = True)
    loss_sum = 0.0
    for i,(datas,labels) in enumerate(train_dataloader):
      #print(i,datas.size(),labels)
      #梯度清零
      optimizer.zero_grad()
      #輸入
      input = V(datas.cuda()).float()
      #目標(biāo)
      target = V(labels.cuda()).long()
      #輸出
      score = model(input).cuda()
      #損失
      loss = criterion(score,target)
      loss_sum += loss
      #反向傳播
      loss.backward()
      #梯度更新
      optimizer.step()      
    print('{}{}{}{}{}'.format('epoch:',epoch,',','loss:',loss))
    torch.save(model.state_dict(),'models/mobilenet_%d.pth'%(epoch+370))

RuntimeError: Expected object of scalar type Long but got scalar type Float for argument #2 'target'

解決方案：target = target.long()

以上這篇pytorch實(shí)現(xiàn)對(duì)輸入超過三通道的數(shù)據(jù)進(jìn)行訓(xùn)練就是小編分享給大家的全部內(nèi)容了，希望能給大家一個(gè)參考，也希望大家多多支持腳本之家。

您可能感興趣的文章: