Image_classification_Pytorch

本来是自己写来比赛用，方便调参的图像分类框架, 一个挺方便分类框架，还在慢慢完善各种模型

目前包含

Resnet18
Resnet50
Se_resnext50
EfficientNetB0
EfficientNetB5
Resnest50
Resnest101

install

git clone https://github.com/Stephenfang51/image_classification_pytorch

requirement

torch 版本必须 >= 1.1 而且必须从 pip3 install https://download.pytorch.org/whl/cu100/torch-1.1.0-cp36-cp36m-linux_x86_64.whl 这样安装 cuda才会是10.0
pip install pretrainedmodels

train data

Preparing your train data like

# if you have 3 classes in total, they are class A, class B, class C
# your train will be like below

train/A/
train/B/
train/C/

if your training image all in one dir, you can use csv_data_classification.ipynb to move your original imgs to its independant dir according to your csv file

How to train

you can directly command like below to train, select your model, input_size, num_classes …etc

python train.py --train_path /input/mango/train_3levels/ --model efficientb5 --input_size 224 --num_classes 3 --lr_scheduler steplr --step_size 20 --batch_size 32 --init_lr 0.001 --freeze True

To see all the command param, you can check train_detail.py

Update

update 2020.4.21

新增ResNest50, ResNest101
- model/utils.py freeze_param, select_network添加ResNest50, ResNest101
- Train_detail —model 添加ResNest50, ResNest101
utils.py 下opcounter 修复格式

update 2020.3.21

改一下train.py (freeze的部分)
将freeze_param添加到model/utils.py底下
data_aug 修正 RandomErasing , 所以dataset.py也修正
train.py / train_detail.py 也修正

update 2020.3.20

dataset.py 修改 (待测试randomErasing)
train.py修改（新增efficientNet, 增加steplr scheduler)
train_detail.py (增加steplr scheduler)
新增models/utils.py (select_networks)
修改utils.py (新增select_scheduler, opcounter)

update 2020.3.16

让argparser support bool值当命令行参数
添加resume/checkpoint参数， resume表示要从checkpoint继续训练， checkpoint 选择要续run的model name，默认从./checkpoints/找
更新test.py 新增test_detail.py吗，主要完成读取模型并且inference后生成csv档
更新dataset中的CDC类，将test_path传入

update 2020.3.13

组建一个图像分类通用框架，用torch搭建

待完善