dataset/pbr.py

import cv2
import torch
from torch.utils.data import Dataset
from torchvision.transforms import Compose

from dataset.transform import Resize, NormalizeImage, PrepareForNet, Crop

class PBRDataset(Dataset):
    def __init__(self, filelist_path, mode, size=(512, 512)):
        self.mode = mode
        self.size = size
        
        # Read filelist using @@ as delimiter
        self.filelist = []
        with open(filelist_path, 'r') as f:
            for line in f:
                line = line.strip()
                # Split on @@ delimiter
                if '@@' in line:  # Use @@ as delimiter between paths
                    parts = line.split('@@')
                    if len(parts) == 2:
                        self.filelist.append((parts[0].strip(), parts[1].strip()))
        
        print(f"Loaded {len(self.filelist)} image pairs")
        
        net_w, net_h = size
        self.transform = Compose([
            Resize(
                width=net_w,
                height=net_h,
                resize_target=True if mode == 'train' else False,
                keep_aspect_ratio=True,
                ensure_multiple_of=12,
                resize_method='lower_bound',
                image_interpolation_method=cv2.INTER_CUBIC,
            ),
            NormalizeImage(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
            PrepareForNet(),
        ] + ([Crop(size[0])] if self.mode == 'train' else []))
        
    def __getitem__(self, item):
        try:
            img_path, disp_path = self.filelist[item]
            
            image = cv2.imread(img_path)
            if image is None:
                print(f"Failed to load image: {img_path}")
                return self.__getitem__((item + 1) % len(self.filelist))
            image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) / 255.0
            
            depth = cv2.imread(disp_path, cv2.IMREAD_GRAYSCALE)
            if depth is None:
                print(f"Failed to load depth: {disp_path}")
                return self.__getitem__((item + 1) % len(self.filelist))
            depth = depth.astype('float32') / 255.0

            sample = self.transform({'image': image, 'depth': depth})

            sample['image'] = torch.from_numpy(sample['image'])
            sample['depth'] = torch.from_numpy(sample['depth'])
            sample['valid_mask'] = torch.ones_like(sample['depth'], dtype=torch.bool)
            sample['image_path'] = img_path
            
            return sample
            
        except Exception as e:
            print(f"Error loading {item}: {str(e)}")
            return self.__getitem__((item + 1) % len(self.filelist))

    def __len__(self):
        return len(self.filelist)