Atlas.py：定义数据集。继承Dataset，用dataloader封装后可以方便地调用图片和标签
utils.py：F1_soft计算F1 score，fit_val计算最适合验证集的阈值
train.py：很简单很常规的训练代码，可以看一下pytorch的训练流程
model.py：定义了用的模型，用的是预训练的resnet18
config.yml：设置训练的参数
test.py：用验证集上得到的阈值，对测试集进行预测，保存的结果submission.csv可以直接提交
任务分配：配置服务器环境（陈）；跑通baseline（勇）；混淆矩阵（年）；结果分析（俞）；设定阈值提交结果（天）；搭建resnet18（张）
https://www.kaggle.com/rejpalcz/cnn-128x128x4-keras-from-scratch-lb-0-328
images by Jonathan Schnabel
green: protein
red: microtubules
blue: nucleus
yellow: endoplasmic reticulum
0 Nucleoplasm
1 Nuclear membrane
2 Nucleoli
3 Nucleoli fibrillar center
4 Nuclear speckles
5 Nuclear bodies
6 Endoplasmic reticulum
7 Golgi apparatus
8 Peroxisomes
9 Endosomes
10 Lysosomes
11 Intermediate filaments
12 Actin filaments
13 Focal adhesion sites
14 Microtubules
15 Microtubule ends
16 Cytokinetic bridge
17 Mitotic spindle
18 Microtubule organizing center
19 Centrosome
20 Lipid droplets
21 Plasma membrane
22 Cell junctions
23 Mitochondria
24 Aggresome
25 Cytosol
26 Cytoplasmic bodies
27 Rods & rings