DeepNoise: Signal and Noise Disentanglement based on Classifying Fluorescent Microscopy Images via Deep Learningg(Genomics, Proteomics and Bioinformatics)

ArXiv | Cite

kaggle NIPS2019 Recursion Cellular Image Classification Challenge 2nd place code

Dependencies

• imgaug == 0.2.8
• opencv-python==3.4.2
• scikit-image==0.14.0
• scikit-learn==0.19.1
• scipy==1.1.0
• torch==1.0.1.
• torchvision==0.2.2

Solution Development

Input and preprocessing:

• 6 channel input, image size 512*512
• per image standardization: normalization of 6 channels in images per plate, with small randomization
• augmentation: random flip, random rotation multiple of 90 degrees.

Backbone CNN:

• Xception and Xception_large;
• Xception is trained from imagenet pretrained weights.
• Xception_large is trained from srcatch which need more trainning epochs.

Loss fuction:

• Arcface Loss s=30, m=0.1:

Path Setup

Set the following path to your own in ./setting.py

data_dir = r'your-own-path/recursion-cellular-image-classification'#data path

Single Model Training

pretrain xception_large 512x512 fold 0 with all cell types：

CUDA_VISIBLE_DEVICES=0,1,2,3 python main.py --mode=pretrain --model=xception_large --image_size=512 --fold_index=0 --batch_size=64

finetune xception_large 512x512 fold 0 on each cell type：

CUDA_VISIBLE_DEVICES=0,1,2,3 python main.py --mode=semi_finetune --model=xception_large --image_size=512 --fold_index=0 --batch_size=64

predict xception_large 512x512 fold 0 model：

CUDA_VISIBLE_DEVICES=0,1,2,3 python main.py --mode=infer --model=xception_large --image_size=512 --fold_index=0 --batch_size=64

generate prediction

python ensemble.py

Reference

If you find our work useful in your research or if you use parts of this code please consider citing our paper:

title = {DeepNoise: Signal and Noise Disentanglement based on Classifying Fluorescent Microscopy Images via Deep Learning},
author={Sen,Yang and Tao,Shen and Yuqi,Fang and Xiyue,Wang and Jun,Zhang and Wei,Yang and Junzhou,Huang and Xiao,Han},
journal = {Genomics, Proteomics and Bioinformatics},  
year = {2023}