How to read images in TIFF format as a dataset.

[o-lucky-o]

I created a dataset called MyDataset, and the input images are (c, h, w). Since c>3, I chose a TIFF format image for my input images.

I learned from #2903 that I can use backend_ args ，this parameter is used to specify the backend for reading images, and supports Tiff format images .

I followed #2468 and mmengine.fileio.io to operate, but an error occurred, so how should I solve this problem.

File "/data/lh/miniconda3/envs/torch/lib/python3.8/site-packages/mmengine/fileio/io.py", line 97, in _get_file_backend backend = backends[backend_name](** backend_args_bak) KeyError: 'tifffile'

During handling of the above exception, another exception occurred: File "/data/lh/miniconda3/envs/torch/lib/python3.8/site-packages/mmengine/registry/build_functions.py", line 144, in build_from_cfg raise type(e)( KeyError: "class MyDataset in mmseg/datasets/MyDataset.py: 'tifffile'"

File "/data/lh/miniconda3/envs/torch/lib/python3.8/site-packages/mmengine/registry/build_functions.py", line 144, in build_from_cfg raise type(e)( KeyError: 'class IterBasedTrainLoop in mmengine/runner/loops.py: "class MyDataset in mmseg/datasets/MyDataset.py: \'tifffile\'"'

from mmseg.registry import DATASETS
from .basesegdataset import BaseSegDataset

@DATASETS.register_module()
class MyDataset(BaseSegDataset):
    # 类别和对应的 RGB配色
    METAINFO = {
        'classes':['blackground', 'haha']],
        'palette':[[255,255,255],[0,255,0]]
    }

    # 指定图像扩展名、标注扩展名
    def __init__(self,
                 img_suffix='.tiff',  # 数据集图片的后缀
                 seg_map_suffix='.png',   # 标注mask图像的后缀格式
                 reduce_zero_label=False, # 类别ID为0的类别是否需要除去
                 **kwargs) -> None:

        super().__init__(
                 img_suffix=img_suffix,
                 seg_map_suffix=seg_map_suffix,
                 reduce_zero_label=reduce_zero_label,
                 backend_args={'backend': 'tifffile'},
                 **kwargs)

[AI-Tianlong]

You can change your dataset configs

train_pipeline = [
    dict(type=LoadSingleRSImagFromFile)

more details about LoadSingleRSImagFromFile in, https://github.com/open-mmlab/mmsegmentation/blob/c685fe6767c4cadf6b051983ca6208f1b9d1ccb8/mmseg/datasets/transforms/loading.py#L503-L556

[jacksteussie]

Contributor

How is this different than using imdecode_backend=tifffile in the 'LoadAnnotation' or 'LoadImageFromFile' transformations?

[AI-Tianlong]

Contributor

How is this different than using imdecode_backend=tifffile in the 'LoadAnnotation' or 'LoadImageFromFile' transformations?

I'm not sure whether backends tifffile support, but if you want to deal with multi-channel remote sensing images, I am more recommend LoadSingleRSImageFromFile, it is based on the gdal backend. You can use conda install gdal to install GDAL, it's works really well.

[AI-Tianlong]

You can change your dataset configs

train_pipeline = [
    dict(type=LoadSingleRSImagFromFile)

more details about LoadSingleRSImagFromFile in, https://github.com/open-mmlab/mmsegmentation/blob/c685fe6767c4cadf6b051983ca6208f1b9d1ccb8/mmseg/datasets/transforms/loading.py#L503-L556

@AI-Tianlong I tried this but got "KeyError: 'LoadSingleRSImagFromFile is not in the mmseg::transform registry. Please check whether the value of LoadSingleRSImagFromFile is correct or it was registered as expected. More details can be found at https://mmengine.readthedocs.io/en/latest/advanced_tutorials/config.html#import-the-custom-module'"

However I can find this function in the mmsegmentation/mmseg/datasets/transforms/loading.py file

my packages` version mmcv 2.1.0 pypi_0 pypi mmengine 0.10.4 pypi_0 pypi mmsegmentation 1.2.2 pypi_0 pypi

Hello，我猜老铁是个Chinese，我用中文说可能表达更清楚。 我这个例子

train_pipeline = [
    dict(type=LoadSingleRSImagFromFile)

是用的 new config，所以没有加引号。 如果你用的是正常的config的话

train_pipeline = [
    dict(type='LoadSingleRSImagFromFile')

要这么写，理论上不会出现这个错误。 你可以再检查一下你的 https://github.com/open-mmlab/mmsegmentation/blob/b040e147adfa027bbc071b624bedf0ae84dfc922/mmseg/datasets/transforms/__init__.py#L27 这个位置，是否有 LoadSingleRSImagFromFile

[crosage]

You can change your dataset configs

train_pipeline = [
    dict(type=LoadSingleRSImagFromFile)

more details about LoadSingleRSImagFromFile in, https://github.com/open-mmlab/mmsegmentation/blob/c685fe6767c4cadf6b051983ca6208f1b9d1ccb8/mmseg/datasets/transforms/loading.py#L503-L556

@AI-Tianlong I tried this but got "KeyError: 'LoadSingleRSImagFromFile is not in the mmseg::transform registry. Please check whether the value of LoadSingleRSImagFromFile is correct or it was registered as expected. More details can be found at https://mmengine.readthedocs.io/en/latest/advanced_tutorials/config.html#import-the-custom-module'" However I can find this function in the mmsegmentation/mmseg/datasets/transforms/loading.py file my packages` version mmcv 2.1.0 pypi_0 pypi mmengine 0.10.4 pypi_0 pypi mmsegmentation 1.2.2 pypi_0 pypi

Hello，我猜老铁是个Chinese，我用中文说可能表达更清楚。 我这个例子

train_pipeline = [
    dict(type=LoadSingleRSImagFromFile)

是用的 new config，所以没有加引号。 如果你用的是正常的config的话

train_pipeline = [
    dict(type='LoadSingleRSImagFromFile')

要这么写，理论上不会出现这个错误。 你可以再检查一下你的

https://github.com/open-mmlab/mmsegmentation/blob/b040e147adfa027bbc071b624bedf0ae84dfc922/mmseg/datasets/transforms/__init__.py#L27

这个位置，是否有 LoadSingleRSImagFromFile

十分感谢，是我拼写有问题