[Feature] Support NYU depth estimation dataset (#3269)

Thanks for your contribution and we appreciate it a lot. The following
instructions would make your pull request more healthy and more easily
get feedback. If you do not understand some items, don't worry, just
make the pull request and seek help from maintainers.

## Motivation

Please describe the motivation of this PR and the goal you want to
achieve through this PR.

## Modification

Please briefly describe what modification is made in this PR.
1. add `NYUDataset`class
2. add script to process NYU dataset
3. add transforms for loading depth map
4. add docs & unittest

## BC-breaking (Optional)

Does the modification introduce changes that break the
backward-compatibility of the downstream repos?
If so, please describe how it breaks the compatibility and how the
downstream projects should modify their code to keep compatibility with
this PR.

## Use cases (Optional)

If this PR introduces a new feature, it is better to list some use cases
here, and update the documentation.

## Checklist

1. Pre-commit or other linting tools are used to fix the potential lint
issues.
5. The modification is covered by complete unit tests. If not, please
add more unit test to ensure the correctness.
6. If the modification has potential influence on downstream projects,
this PR should be tested with downstream projects, like MMDet or
MMDet3D.
7. The documentation has been modified accordingly, like docstring or
example tutorials.

docs/en/user_guides/2_dataset_prepare.md

@@ -198,6 +198,13 @@ mmsegmentation
 |   │   │   │   └── rles
 |   │   │   │   │   ├──sem_seg_train.json
 |   │   │   │   │   └──sem_seg_val.json
+│   ├── nyu
+│   │   ├── images
+│   │   │   ├── train
+│   │   │   ├── test
+│   │   ├── annotations
+│   │   │   ├── train
+│   │   │   ├── test
 ```
 
 ## Download dataset via MIM
@@ -735,3 +742,13 @@ mmsegmentation
 |   │   │   │   │   ├──sem_seg_train.json
 |   │   │   │   │   └──sem_seg_val.json
 ```
+
+## NYU
+
+- To access the NYU dataset, you can download it from [this link](https://drive.google.com/file/d/1wC-io-14RCIL4XTUrQLk6lBqU2AexLVp/view?usp=share_link)
+
+- Once the download is complete, you can utilize the [tools/dataset_converters/nyu.py](/tools/dataset_converters/nyu.py) script to extract and organize the data into the required format. Run the following command in your terminal:
+
+  ```bash
+  python tools/dataset_converters/nyu.py nyu.zip
+  ```

docs/zh_cn/user_guides/2_dataset_prepare.md

@@ -198,6 +198,13 @@ mmsegmentation
 |   │   │   │   └── rles
 |   │   │   │   │   ├──sem_seg_train.json
 |   │   │   │   │   └──sem_seg_val.json
+│   ├── nyu
+│   │   ├── images
+│   │   │   ├── train
+│   │   │   ├── test
+│   │   ├── annotations
+│   │   │   ├── train
+│   │   │   ├── test
 ```
 
 ## 用 MIM 下载数据集
@@ -731,3 +738,13 @@ mmsegmentation
 |   │   │   │   │   ├──sem_seg_train.json
 |   │   │   │   │   └──sem_seg_val.json
 ```
+
+## NYU
+
+- 您可以从 [这个链接](https://drive.google.com/file/d/1wC-io-14RCIL4XTUrQLk6lBqU2AexLVp/view?usp=share_link) 下载 NYU 数据集
+
+- 下载完成后，您可以使用 [tools/dataset_converters/nyu.py](/tools/dataset_converters/nyu.py) 脚本来解压和组织数据到所需的格式
+
+  ```bash
+  python tools/dataset_converters/nyu.py nyu.zip
+  ```

mmseg/datasets/__init__.py

@@ -19,6 +19,7 @@
 from .loveda import LoveDADataset
 from .mapillary import MapillaryDataset_v1, MapillaryDataset_v2
 from .night_driving import NightDrivingDataset
+from .nyu import NYUDataset
 from .pascal_context import PascalContextDataset, PascalContextDataset59
 from .potsdam import PotsdamDataset
 from .refuge import REFUGEDataset
@@ -58,5 +59,6 @@
     'SynapseDataset', 'REFUGEDataset', 'MapillaryDataset_v1',
     'MapillaryDataset_v2', 'Albu', 'LEVIRCDDataset',
     'LoadMultipleRSImageFromFile', 'LoadSingleRSImageFromFile',
-    'ConcatCDInput', 'BaseCDDataset', 'DSDLSegDataset', 'BDD100KDataset'
+    'ConcatCDInput', 'BaseCDDataset', 'DSDLSegDataset', 'BDD100KDataset',
+    'NYUDataset'
 ]

mmseg/datasets/nyu.py

@@ -0,0 +1,123 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+import os.path as osp
+from typing import List
+
+import mmengine.fileio as fileio
+
+from mmseg.registry import DATASETS
+from .basesegdataset import BaseSegDataset
+
+
+@DATASETS.register_module()
+class NYUDataset(BaseSegDataset):
+    """NYU depth estimation dataset. The file structure should be.
+
+    .. code-block:: none
+
+        ├── data
+        │   ├── nyu
+        │   │   ├── images
+        │   │   │   ├── train
+        │   │   │   │   ├── scene_xxx.jpg
+        │   │   │   │   ├── ...
+        │   │   │   ├── test
+        │   │   ├── annotations
+        │   │   │   ├── train
+        │   │   │   │   ├── scene_xxx.png
+        │   │   │   │   ├── ...
+        │   │   │   ├── test
+
+    Args:
+        ann_file (str): Annotation file path. Defaults to ''.
+        metainfo (dict, optional): Meta information for dataset, such as
+            specify classes to load. Defaults to None.
+        data_root (str, optional): The root directory for ``data_prefix`` and
+            ``ann_file``. Defaults to None.
+        data_prefix (dict, optional): Prefix for training data. Defaults to
+            dict(img_path='images', depth_map_path='annotations').
+        img_suffix (str): Suffix of images. Default: '.jpg'
+        seg_map_suffix (str): Suffix of segmentation maps. Default: '.png'
+        filter_cfg (dict, optional): Config for filter data. Defaults to None.
+        indices (int or Sequence[int], optional): Support using first few
+            data in annotation file to facilitate training/testing on a smaller
+            dataset. Defaults to None which means using all ``data_infos``.
+        serialize_data (bool, optional): Whether to hold memory using
+            serialized objects, when enabled, data loader workers can use
+            shared RAM from master process instead of making a copy. Defaults
+            to True.
+        pipeline (list, optional): Processing pipeline. Defaults to [].
+        test_mode (bool, optional): ``test_mode=True`` means in test phase.
+            Defaults to False.
+        lazy_init (bool, optional): Whether to load annotation during
+            instantiation. In some cases, such as visualization, only the meta
+            information of the dataset is needed, which is not necessary to
+            load annotation file. ``Basedataset`` can skip load annotations to
+            save time by set ``lazy_init=True``. Defaults to False.
+        max_refetch (int, optional): If ``Basedataset.prepare_data`` get a
+            None img. The maximum extra number of cycles to get a valid
+            image. Defaults to 1000.
+        ignore_index (int): The label index to be ignored. Default: 255
+        reduce_zero_label (bool): Whether to mark label zero as ignored.
+            Default to False.
+        backend_args (dict, Optional): Arguments to instantiate a file backend.
+            See https://mmengine.readthedocs.io/en/latest/api/fileio.htm
+            for details. Defaults to None.
+            Notes: mmcv>=2.0.0rc4, mmengine>=0.2.0 required.
+    """
+    METAINFO = dict(
+        classes=('printer_room', 'bathroom', 'living_room', 'study',
+                 'conference_room', 'study_room', 'kitchen', 'home_office',
+                 'bedroom', 'dinette', 'playroom', 'indoor_balcony',
+                 'laundry_room', 'basement', 'excercise_room', 'foyer',
+                 'home_storage', 'cafe', 'furniture_store', 'office_kitchen',
+                 'student_lounge', 'dining_room', 'reception_room',
+                 'computer_lab', 'classroom', 'office', 'bookstore'))
+
+    def __init__(self,
+                 data_prefix=dict(
+                     img_path='images', depth_map_path='annotations'),
+                 img_suffix='.jpg',
+                 depth_map_suffix='.png',
+                 **kwargs) -> None:
+        super().__init__(
+            data_prefix=data_prefix,
+            img_suffix=img_suffix,
+            seg_map_suffix=depth_map_suffix,
+            **kwargs)
+
+    def _get_category_id_from_filename(self, image_fname: str) -> int:
+        """Retrieve the category ID from the given image filename."""
+        image_fname = osp.basename(image_fname)
+        position = image_fname.find(next(filter(str.isdigit, image_fname)), 0)
+        categoty_name = image_fname[:position - 1]
+        if categoty_name not in self._metainfo['classes']:
+            return -1
+        else:
+            return self._metainfo['classes'].index(categoty_name)
+
+    def load_data_list(self) -> List[dict]:
+        """Load annotation from directory or annotation file.
+
+        Returns:
+            list[dict]: All data info of dataset.
+        """
+        data_list = []
+        img_dir = self.data_prefix.get('img_path', None)
+        ann_dir = self.data_prefix.get('depth_map_path', None)
+
+        _suffix_len = len(self.img_suffix)
+        for img in fileio.list_dir_or_file(
+                dir_path=img_dir,
+                list_dir=False,
+                suffix=self.img_suffix,
+                recursive=True,
+                backend_args=self.backend_args):
+            data_info = dict(img_path=osp.join(img_dir, img))
+            if ann_dir is not None:
+                depth_map = img[:-_suffix_len] + self.seg_map_suffix
+                data_info['depth_map_path'] = osp.join(ann_dir, depth_map)
+            data_info['seg_fields'] = []
+            data_info['category_id'] = self._get_category_id_from_filename(img)
+            data_list.append(data_info)
+        data_list = sorted(data_list, key=lambda x: x['img_path'])
+        return data_list

mmseg/datasets/transforms/__init__.py

@@ -2,8 +2,8 @@
 from .formatting import PackSegInputs
 from .loading import (LoadAnnotations, LoadBiomedicalAnnotation,
                       LoadBiomedicalData, LoadBiomedicalImageFromFile,
-                      LoadImageFromNDArray, LoadMultipleRSImageFromFile,
-                      LoadSingleRSImageFromFile)
+                      LoadDepthAnnotation, LoadImageFromNDArray,
+                      LoadMultipleRSImageFromFile, LoadSingleRSImageFromFile)
 # yapf: disable
 from .transforms import (CLAHE, AdjustGamma, Albu, BioMedical3DPad,
                          BioMedical3DRandomCrop, BioMedical3DRandomFlip,
@@ -24,5 +24,5 @@
     'ResizeShortestEdge', 'BioMedicalGaussianNoise', 'BioMedicalGaussianBlur',
     'BioMedical3DRandomFlip', 'BioMedicalRandomGamma', 'BioMedical3DPad',
     'RandomRotFlip', 'Albu', 'LoadSingleRSImageFromFile', 'ConcatCDInput',
-    'LoadMultipleRSImageFromFile'
+    'LoadMultipleRSImageFromFile', 'LoadDepthAnnotation'
 ]

mmseg/datasets/transforms/formatting.py

@@ -92,6 +92,11 @@ def transform(self, results: dict) -> dict:
                                                       ...].astype(np.int64)))
             data_sample.set_data(dict(gt_edge_map=PixelData(**gt_edge_data)))
 
+        if 'gt_depth_map' in results:
+            gt_depth_data = dict(
+                data=to_tensor(results['gt_depth_map'][None, ...]))
+            data_sample.set_data(dict(gt_depth_map=PixelData(**gt_depth_data)))
+
         img_meta = {}
         for key in self.meta_keys:
             if key in results:

mmseg/datasets/transforms/loading.py

@@ -625,3 +625,77 @@ def __repr__(self):
         repr_str = (f'{self.__class__.__name__}('
                     f'to_float32={self.to_float32})')
         return repr_str
+
+
+@TRANSFORMS.register_module()
+class LoadDepthAnnotation(BaseTransform):
+    """Load ``depth_map`` annotation provided by depth estimation dataset.
+
+    The annotation format is as the following:
+
+    .. code-block:: python
+
+        {
+            'gt_depth_map': np.ndarray [Y, X]
+        }
+
+    Required Keys:
+
+    - seg_depth_path
+
+    Added Keys:
+
+    - gt_depth_map (np.ndarray): Depth map with shape (Y, X) by
+        default, and data type is float32 if set to_float32 = True.
+
+    Args:
+        decode_backend (str): The data decoding backend type. Options are
+            'numpy', 'nifti', and 'cv2'. Defaults to 'cv2'.
+        to_float32 (bool): Whether to convert the loaded depth map to a float32
+            numpy array. If set to False, the loaded image is an uint16 array.
+            Defaults to True.
+        depth_rescale_factor (float): Factor to rescale the depth value to
+            limit the range. Defaults to 1.0.
+        backend_args (dict, Optional): Arguments to instantiate a file backend.
+            See :class:`mmengine.fileio` for details.
+            Defaults to None.
+            Notes: mmcv>=2.0.0rc4, mmengine>=0.2.0 required.
+    """
+
+    def __init__(self,
+                 decode_backend: str = 'cv2',
+                 to_float32: bool = True,
+                 depth_rescale_factor: float = 1.0,
+                 backend_args: Optional[dict] = None) -> None:
+        super().__init__()
+        self.decode_backend = decode_backend
+        self.to_float32 = to_float32
+        self.depth_rescale_factor = depth_rescale_factor
+        self.backend_args = backend_args.copy() if backend_args else None
+
+    def transform(self, results: Dict) -> Dict:
+        """Functions to load depth map.
+
+        Args:
+            results (dict): Result dict from :obj:``mmcv.BaseDataset``.
+
+        Returns:
+            dict: The dict contains loaded depth map.
+        """
+        data_bytes = fileio.get(results['depth_map_path'], self.backend_args)
+        gt_depth_map = datafrombytes(data_bytes, backend=self.decode_backend)
+
+        if self.to_float32:
+            gt_depth_map = gt_depth_map.astype(np.float32)
+
+        gt_depth_map *= self.depth_rescale_factor
+        results['gt_depth_map'] = gt_depth_map
+        results['seg_fields'].append('gt_depth_map')
+        return results
+
+    def __repr__(self):
+        repr_str = (f'{self.__class__.__name__}('
+                    f"decode_backend='{self.decode_backend}', "
+                    f'to_float32={self.to_float32}, '
+                    f'backend_args={self.backend_args})')
+        return repr_str

mmseg/utils/io.py

@@ -3,6 +3,7 @@
 import io
 import pickle
 
+import cv2
 import numpy as np
 
 
@@ -12,7 +13,7 @@ def datafrombytes(content: bytes, backend: str = 'numpy') -> np.ndarray:
     Args:
         content (bytes): The data bytes got from files or other streams.
         backend (str): The data decoding backend type. Options are 'numpy',
-            'nifti' and 'pickle'. Defaults to 'numpy'.
+            'nifti', 'cv2' and 'pickle'. Defaults to 'numpy'.
 
     Returns:
         numpy.ndarray: Loaded data array.
@@ -33,6 +34,9 @@ def datafrombytes(content: bytes, backend: str = 'numpy') -> np.ndarray:
                 data = Nifti1Image.from_bytes(data.to_bytes()).get_fdata()
             elif backend == 'numpy':
                 data = np.load(f)
+            elif backend == 'cv2':
+                data = np.frombuffer(f.read(), dtype=np.uint16)
+                data = cv2.imdecode(data, 2)
             else:
                 raise ValueError
     return data

tests/test_datasets/test_dataset.py

@@ -9,7 +9,7 @@
                             CityscapesDataset, COCOStuffDataset,
                             DecathlonDataset, DSDLSegDataset, ISPRSDataset,
                             LIPDataset, LoveDADataset, MapillaryDataset_v1,
-                            MapillaryDataset_v2, PascalVOCDataset,
+                            MapillaryDataset_v2, NYUDataset, PascalVOCDataset,
                             PotsdamDataset, REFUGEDataset, SynapseDataset,
                             iSAIDDataset)
 from mmseg.registry import DATASETS
@@ -462,3 +462,14 @@ def test_dsdlseg_dataset():
         assert len(dataset.metainfo['classes']) == 21
     else:
         ImportWarning('Package `dsdl` is not installed.')
+
+
+def test_nyu_dataset():
+    dataset = NYUDataset(
+        data_root='tests/data/pseudo_nyu_dataset',
+        data_prefix=dict(img_path='images', depth_map_path='annotations'),
+    )
+    assert len(dataset) == 1
+    data = dataset[0]
+    assert data.get('depth_map_path', None) is not None
+    assert data.get('category_id', -1) == 26

tests/test_datasets/test_loading.py

@@ -7,10 +7,11 @@
 import numpy as np
 from mmcv.transforms import LoadImageFromFile
 
-from mmseg.datasets.transforms import (LoadAnnotations,
-                                       LoadBiomedicalAnnotation,
+from mmseg.datasets.transforms import LoadAnnotations  # noqa
+from mmseg.datasets.transforms import (LoadBiomedicalAnnotation,
                                        LoadBiomedicalData,
                                        LoadBiomedicalImageFromFile,
+                                       LoadDepthAnnotation,
                                        LoadImageFromNDArray)
 
 
@@ -276,3 +277,19 @@ def test_load_biomedical_data(self):
                                    "decode_backend='numpy', "
                                    'to_xyz=False, '
                                    'backend_args=None)')
+
+    def test_load_depth_annotation(self):
+        input_results = dict(
+            img_path='tests/data/pseudo_nyu_dataset/images/'
+            'bookstore_0001d_00001.jpg',
+            depth_map_path='tests/data/pseudo_nyu_dataset/'
+            'annotations/bookstore_0001d_00001.png',
+            category_id=-1,
+            seg_fields=[])
+        transform = LoadDepthAnnotation(depth_rescale_factor=0.001)
+        results = transform(input_results)
+        assert 'gt_depth_map' in results
+        assert results['gt_depth_map'].shape[:2] == mmcv.imread(
+            input_results['depth_map_path']).shape[:2]
+        assert results['gt_depth_map'].dtype == np.float32
+        assert 'gt_depth_map' in results['seg_fields']