kaggle-classifer's Introduction

Plant Pathology 2021 - FGVC8

Identify the category of foliar diseases in apple trees

Download Dataset

BaiDu Network Disk

passward kwj4

if you have downloaded the dataset,the document structure should be as follows:

../
    ./train_images
    ./test_images
    train.csv
    sample_submission.csv

where ../ is ROOT in config.py

Sample Data and EDA

Data

There are some different classes of image as below:

EDA

Step 1 (Key Step)

Another Step

SoftWare Version and Installation

Below are the versions of opencv-python, torch, torchvision, rich, warmup_scheduler, and numpy currently running at the time of writing this:

opencv-python : 4.5.5.64
torch : 1.10.0+cu102
torchvision : 0.11.0+cu102
rich : 10.16.2
warmup_scheduler : 0.3.2
numpy : 1.20.0

Install rich accelerate DataLoaderX albumentations and warmup_scheduler

pip install git+https://github.com/ildoonet/pytorch-gradual-warmup-lr.git
pip install rich
pip install accelerate
pip install prefetch_generator 
pip install -U albumentations
sudo apt-get install libturbojpeg
pip install -U jpeg4py

or you can use install.sh to configure

Custom Model

Add new model file to ./models/

example

./models/new_model.py

import torch
import torch.nn as nn
class Model(nn.Module):
    def __init__(self,in_channle,num_classes):
        super().__init__()
        pass
    def forward(self):
        pass

Custom transform

Add code in vision_transform

Configure parameters

In config.py,you can see a dict as follows:

parameter = {
    "ROOT": r"/mnt/d/dataset",
    "dataset_csv": r"train.csv",
    "batch_size": 32,
    "num_workers" : 8,
    "learning_rate" : 1e-3,
    "val_size" : 0.2,
    "seed" : 12,
    "epochs":10,
    "num_classes" : 12,
    "in_channel" : 3,
    "image_size" : (128,128)
}

Recommend Projects