Skip to content

PanJinquan/pytorch-learning-notes

Repository files navigation

pytorch-learning-tutorials

觉得可以,麻烦给个”Star“

1.image_classification 图像分类

《pytorch实现L2和L1正则化的方法》https://panjinquan.blog.csdn.net/article/details/88426648

2.object_detection目标检测

mobileNet v1 v2 SSD目标检测:该项目是参考《pytorch-ssd》https://github.com/qfgaohao/pytorch-ssd ,修改,主要是方便训练。
<<<<<<< HEAD 数据集VOC2007和VOC2012:
训练方法:my_train_ssd.py
修改my_train_ssd.py的参数train_filename和val_filename即可,直接运行训练,例如:
======= 数据集VOC2007和VOC2012:
训练方法:my_train_ssd.py
修改my_train_ssd.py的参数train_filename和val_filename即可,直接运行训练,例如:

166dd1f31f4d3c4ce73f13077cf6619bbff91635

train_filename = 'E:/git/VOC0712_dataset/train.txt' #训练文件
val_filename = 'E:/git/VOC0712_dataset/val.txt'     #测试文件

测试方法:run_ssd_example.py

net_type = 'mb2-ssd-lite' #模型类型
model_path = 'models/mb2-ssd-lite-Epoch-190-Loss-3.0529016691904802.pth'#模型路径
label_path = 'models/voc-model-labels.txt'#label文件路径
image_path = './dataset/images/6.jpg'#测试图片

3.DeepLearningTutorials教程

网上收集的Pytorch的学习资料

4.caffe2-android

在android上运行caffe2模型实现图像识别的demo

5.UNet图像分割

使用UNet模型实现的图像分割

一点笔记

<<<<<<< HEAD

nohup:服务器后台训练,并打印log

nohup python train.py --batch_size=4 1>> train.log &

查看进程并杀死某个进程

jobs -l
ps aux|grep python
kill 7080

=======

166dd1f31f4d3c4ce73f13077cf6619bbff91635 查看GPU使用情况

nvidia-smi

指定使用GPU:

import os
os.environ["CUDA_VISIBLE_DEVICES"] = "1"#编号从0开始

pytorch版本,检查GPU是否可用

import torch
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# device='cuda'
print("-----device:{}".format(device))
print("-----Pytorch version:{}".format(torch.__version__))

4.相关说明

Pytorch version:1.0.0

About

pytorch learning tutorials

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published