What causes "image: {} isn't RGB mode." in WZMIAOMIAO/deep-learning-for-image-processing?

Change the loader to img.convert('RGB') instead of raising

WZMIAOMIAO/deep-learning-for-image-processing · error · ValueError

image: {} isn't RGB mode.

Error message

image: {} isn't RGB mode.

What it means

Custom dataset's MyDataSet.__getitem__ enforces PIL mode 'RGB' for every sample because the classification/transform pipeline assumes 3-channel images. Any grayscale, palette, RGBA, or CMYK image in the dataset folders raises ValueError listing the offending file path.

Source

Thrown at pytorch_classification/custom_dataset/my_dataset.py:21

from torch.utils.data import Dataset


class MyDataSet(Dataset):
    """自定义数据集"""

    def __init__(self, images_path: list, images_class: list, transform=None):
        self.images_path = images_path
        self.images_class = images_class
        self.transform = transform

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

    def __getitem__(self, item):
        img = Image.open(self.images_path[item])
        # RGB为彩色图片，L为灰度图片
        if img.mode != 'RGB':
            raise ValueError("image: {} isn't RGB mode.".format(self.images_path[item]))
        label = self.images_class[item]

        if self.transform is not None:
            img = self.transform(img)

        return img, label

    @staticmethod
    def collate_fn(batch):
        # 官方实现的default_collate可以参考
        # https://github.com/pytorch/pytorch/blob/67b7e751e6b5931a9f45274653f4f653a4e6cdf6/torch/utils/data/_utils/collate.py
        images, labels = tuple(zip(*batch))

        images = torch.stack(images, dim=0)
        labels = torch.as_tensor(labels)
        return images, labels

View on GitHub (pinned to 1ec3fe6f37)

Solutions

Change the loader to img.convert('RGB') instead of raising
Batch-normalize the dataset offline to RGB mode (PIL script or ImageMagick)
Filter/clean the dataset: enumerate files, open each, and convert or drop non-RGB images before training

Example fix

// before
if img.mode != 'RGB':
    raise ValueError("image: {} isn't RGB mode.".format(self.images_path[item]))
// after
if img.mode != 'RGB':
    print("converted", self.images_path[item])
    img = img.convert('RGB')

Defensive patterns

Strategy: validation

Validate before calling

from PIL import Image
from pathlib import Path
bad = [p for p in Path(data_root).rglob('*.jpg') if Image.open(p).mode != 'RGB']
print(len(bad), 'non-RGB images:', bad[:10])

Type guard

def is_rgb_image(img) -> bool:
    return img.mode == 'RGB'

Try / catch

try:
    img, label = dataset[i]
except ValueError as e:
    if "isn't RGB mode" in str(e):
        path = str(e).split('image: ')[1].split(' ')[0]
        img, label = Image.open(path).convert('RGB'), dataset.images_class[i]
    else:
        raise

Prevention

Clean custom datasets on ingestion: convert all images to RGB at copy time
Add mode validation to your dataset-preparation checklist
Prefer raising-with-path behavior locally to find files, but fix them before distributed training

When it happens

Trigger: Indexing the custom dataset (via DataLoader or direct ds[i]) where images_path[item] points to a non-RGB-mode file, e.g. an 'L' grayscale photo or 'P'-mode PNG.

Common situations: User-provided custom image folders containing grayscale phone scans, RGBA PNG logos, or palette-mode images; no offline normalization of the dataset before training.

Related errors

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/215be9eaf74f7f05. Report an issue: GitHub.