S2_TimberDataset.py

import torch
from torch import Tensor
from torchvision import transforms
import cv2
from torch.utils.data import Dataset, DataLoader
import os
from PIL import Image
import pandas as pd
# 3264 x 2448

DATA_DIR = "data/image/train"
labels = os.listdir(DATA_DIR)
label2id = {label:id for id, label in enumerate(labels)}

def compile_image_df(data_dir:str, split_at = 0.9)-> pd.DataFrame:
    dirs = os.listdir(data_dir)
    columns=['Image_ID','Species']
    train = pd.DataFrame(columns=columns)
    val = pd.DataFrame(columns=columns)
    for dir in dirs:
        imgs = [(f"{data_dir}/{dir}/{img}", dir) for img in list(os.listdir(f"{data_dir}/{dir}"))]
        length = len(imgs)
        train_count = int(length * split_at)
        train = pd.concat([train, pd.DataFrame(imgs[:train_count],columns=columns)])
        val = pd.concat([val, pd.DataFrame(imgs[train_count:],columns=columns)])

    return train, val

class TimberDataset(Dataset):
    def __init__(self, 
                 dataframe: pd.DataFrame, 
                 is_train=False, 
                 transform=None,
                 device=torch.device('cuda' if torch.cuda.is_available() else 'cpu')) -> None:
        super().__init__()
        self.dataframe = dataframe
        self.is_train = is_train
        self.transform = transform
        self.device = device

    def __len__(self) -> int:
        return len(self.dataframe)

    def __getitem__(self, idx: list[int]|Tensor):
        if torch.is_tensor(idx): 
            idx = idx.tolist()
        
        img_name = os.path.join(self.dataframe.iloc[idx,0])
        image = cv2.imread(img_name)
        image = Image.fromarray(image)

        label = self.dataframe.iloc[idx,1]
        label = label2id[label]
        label = torch.tensor(int(label))

        if self.transform:
            image = self.transform(image)
        return image.to(self.device), label.to(self.device)

def build_dataloader(
        train_ratio = 0.9,
        img_size = (640,640),
        batch_size = 12,
    ) -> tuple[DataLoader,DataLoader]:
    train_df, val_df = compile_image_df(DATA_DIR, split_at=train_ratio)

    transform = transforms.Compose([
        transforms.Resize(img_size),
        transforms.ToTensor(),
    ])

    train_loader = DataLoader(TimberDataset(train_df, is_train=True,transform=transform),
                              shuffle=True,
                              batch_size=batch_size)
    val_loader = DataLoader(TimberDataset(val_df, is_train=True,transform=transform),
                            batch_size=batch_size)
    
    return train_loader,val_loader