added hkh dataset datamodule and download script

README.md

@@ -34,13 +34,15 @@ pip install 'git+https://github.com/isaaccorley/torchrs.git#egg=torch-rs[train]'
 
 * [PROBA-V Multi-Image Super Resolution](https://github.com/isaaccorley/torchrs#proba-v-super-resolution)
 * [ETCI 2021 Flood Detection](https://github.com/isaaccorley/torchrs#etci-2021-flood-detection)
+* [HKH Glacier Mapping](https://github.com/isaaccorley/torchrs#hkh-glacier-mapping)
+* [ZueriCrop - Time-Series Instance Segmentation](https://github.com/isaaccorley/torchrs#zuericrop)
 * [FAIR1M - Fine-grained Object Recognition](https://github.com/isaaccorley/torchrs#fair1m---fine-grained-object-recognition)
 * [ADVANCE - Audiovisual Aerial Scene Recognition](https://github.com/isaaccorley/torchrs#advance---audiovisual-aerial-scene-recognition)
 * [OSCD - Onera Satellite Change Detection](https://github.com/isaaccorley/torchrs#onera-satellite-change-detection-oscd)
 * [S2Looking - Satellite Side-Looking Change Detection](https://github.com/isaaccorley/torchrs#satellite-side-looking-s2looking-change-detection)
 * [LEVIR-CD+ - LEVIR Change Detection+](https://github.com/isaaccorley/torchrs#levir-change-detection-levir-cd)
 * [HRSCD - High Resolution Semantic Change Detection](https://github.com/isaaccorley/torchrs#high-resolution-semantic-change-detection-hrscd)
-* [S2MTCP - Sentinel-2 Multitemporal Cities Pairs Change Detection](https://github.com/isaaccorley/torchrs#sentinel-2-multitemporal-cities-pairs-s2mtcp)
+* [S2MTCP - Sentinel-2 Multitemporal Cities Pairs](https://github.com/isaaccorley/torchrs#sentinel-2-multitemporal-cities-pairs-s2mtcp)
 * [RSVQA LR - Remote Sensing Visual Question Answering Low Resolution](https://github.com/isaaccorley/torchrs#remote-sensing-visual-question-answering-rsvqa-low-resolution-lr)
 * [RSVQAxBEN - Remote Sensing Visual Question Answering BigEarthNet](https://github.com/isaaccorley/torchrs#remote-sensing-visual-question-answering-bigearthnet-rsvqaxben)
 * [RSICD - Remote Sensing Image Captioning Dataset](https://github.com/isaaccorley/torchrs#remote-sensing-image-captioning-dataset-rsicd)
@@ -53,7 +55,6 @@ pip install 'git+https://github.com/isaaccorley/torchrs.git#egg=torch-rs[train]'
 * [Inria Aerial Image Labeling - Building Semantic Segmentation](https://github.com/isaaccorley/torchrs#inria-aerial-image-labeling)
 * [Dubai - Semantic Segmentation](https://github.com/isaaccorley/torchrs#dubai-segmentation)
 * [GID-15 - Semantic Segmentation](https://github.com/isaaccorley/torchrs#gid-15)
-* [ZueriCrop - Time-Series Instance Segmentation](https://github.com/isaaccorley/torchrs#zuericrop)
 * [TiSeLaC - Time-Series Land Cover Classification](https://github.com/isaaccorley/torchrs#tiselac)
 
 ### PROBA-V Super Resolution
@@ -121,6 +122,77 @@ x: dict(
 """
 ```
 
+### HKH Glacier Mapping
+
+<img src="./assets/hkh_glacier.png" width="400px"></img>
+
+The [Hindu Kush Himalayas (HKH) Glacier Mapping](https://lila.science/datasets/hkh-glacier-mapping) dataset is a semantic segmentation dataset of 7,095 512x512 multispectral images taken by the [USGS LandSat 7 satellite](https://landsat.gsfc.nasa.gov/landsat-7). The dataset contains imagery from 2002-2008 of the HKH region (spanning 8 countries) along with separate masks of clean-iced and debris-covered glaciers. The imagery contains 15 bands which includes 10 LandSat 7 bands, 3 precomputed NVDI/NDSI/NDWI indices, and 2 digital elevation and slope maps from the [SRTM 90m DEM Digital Elevation Database](https://srtm.csi.cgiar.org/).
+
+The dataset can be downloaded (18GB/109GB compressed/uncompressed) using `scripts/download_hkh_glacier.sh` and instantiated below:
+
+```python
+from torchrs.transforms import Compose, ToTensor
+from torchrs.datasets import HKHGlacierMapping
+
+transform = Compose([ToTensor()])
+
+dataset = HKHGlacierMapping(
+    root="path/to/dataset/",
+    transform=transform
+)
+
+x = dataset[0]
+"""
+x: dict(
+    x:                   (15, 512, 512)
+    clean_ice_mask:      (1, 512, 512)
+    debris_covered_mask: (1, 256, 256)
+)
+"""
+
+dataset.bands
+"""
+['LE7 B1 (blue)', 'LE7 B2 (green)', 'LE7 B3 (red)', 'LE7 B4 (near infrared)', 'LE7 B5 (shortwave infrared 1)',
+'LE7 B6_VCID_1 (low-gain thermal infrared)', 'LE7 B6_VCID_2 (high-gain thermal infrared)',
+'LE7 B7 (shortwave infrared 2)', 'LE7 B8 (panchromatic)', 'LE7 BQA (quality bitmask)', 'NDVI (vegetation index)',
+'NDSI (snow index)', 'NDWI (water index)', 'SRTM 90 elevation', 'SRTM 90 slope']
+"""
+```
+
+### ZueriCrop
+
+<img src="./assets/zuericrop.png" width="650px"></img>
+
+The [ZueriCrop](https://github.com/0zgur0/ms-convSTAR) dataset is a time-series instance segmentation dataset proposed in ["Crop mapping from image time series: deep learning with multi-scale label hierarchies", Turkoglu et al.](https://arxiv.org/abs/2102.08820) of 116k medium resolution (10m) 24x24 multispectral 9-band imagery of Zurich and Thurgau, Switzerland taken by the [ESA Sentinel-2 satellite](https://sentinel.esa.int/web/sentinel/missions/sentinel-2) and contains pixel level semantic and instance annotations for 48 fine-grained, hierarchical categories of crop types. Note that there is only a single ground truth semantic & instance mask per time-series.
+
+The dataset can be downloaded (39GB) using `scripts/download_zuericrop.sh` and instantiated below:
+
+```python
+from torchrs.transforms import Compose, ToTensor
+from torchrs.datasets import ZueriCrop
+
+transform = Compose([ToTensor()])
+
+dataset = ZueriCrop(
+    root="path/to/dataset/",
+    transform=transform
+)
+
+x = dataset[0]
+"""
+x: dict(
+    x:              (142, 9, 24, 24)    (t, c, h, w)
+    mask:           (1, 24, 24)
+    instance_mask:  (1, 24, 24)
+)
+"""
+
+[cls.label for cls in ds.classes]
+"""
+['Unknown', 'SummerBarley', 'WinterBarley', 'Oat', 'Wheat', 'Grain', ...]
+"""
+```
+
 ### FAIR1M - Fine-grained Object Recognition
 
 <img src="./assets/fair1m.jpg" width="550px"></img>
@@ -747,40 +819,6 @@ dataset.classes
 """
 ```
 
-### ZueriCrop
-
-<img src="./assets/zuericrop.png" width="750px"></img>
-
-The [ZueriCrop](https://github.com/0zgur0/ms-convSTAR) dataset is a time-series instance segmentation dataset proposed in ["Crop mapping from image time series: deep learning with multi-scale label hierarchies", Turkoglu et al.](https://arxiv.org/abs/2102.08820) of 116k medium resolution (10m) 24x24 multispectral 9-band imagery of Zurich and Thurgau, Switzerland taken by the [ESA Sentinel-2 satellite](https://sentinel.esa.int/web/sentinel/missions/sentinel-2) and contains pixel level semantic and instance annotations for 48 fine-grained, hierarchical categories of crop types. Note that there is only a single ground truth semantic & instance mask per time-series.
-
-The dataset can be downloaded (39GB) using `scripts/download_zuericrop.sh` and instantiated below:
-
-```python
-from torchrs.transforms import Compose, ToTensor
-from torchrs.datasets import ZueriCrop
-
-transform = Compose([ToTensor()])
-
-dataset = ZueriCrop(
-    root="path/to/dataset/",
-    transform=transform
-)
-
-x = dataset[0]
-"""
-x: dict(
-    x:              (142, 9, 24, 24)    (t, c, h, w)
-    mask:           (1, 24, 24)
-    instance_mask:  (1, 24, 24)
-)
-"""
-
-[cls.label for cls in ds.classes]
-"""
-['Unknown', 'SummerBarley', 'WinterBarley', 'Oat', 'Wheat', 'Grain', ...]
-"""
-```
-
 ### TiSeLaC
 
 <img src="./assets/tiselac.png" width="900px"></img>

scripts/download_hkh_glacier.sh

@@ -0,0 +1,5 @@
+pip install gdown
+mkdir -p .data
+gdown --id 1cvjfe_MZJI9HXwkRgoQbSCH55qQd6Esm
+unzip hkh_glacier_mapping.zip -d .data/
+rm hkh_glacier_mapping.zip

torchrs/datasets/__init__.py

@@ -21,11 +21,13 @@
 from .zuericrop import ZueriCrop
 from .aid import AID
 from .dubai_segmentation import DubaiSegmentation
+from .hkh_glacier import HKHGlacierMapping
 
 
 __all__ = [
     "PROBAV", "ETCI2021", "RSVQALR", "RSVQAxBEN", "EuroSATRGB", "EuroSATMS",
     "RESISC45", "RSICD", "OSCD", "S2Looking", "LEVIRCDPlus", "FAIR1M",
     "SydneyCaptions", "UCMCaptions", "S2MTCP", "ADVANCE", "SAT4", "SAT6",
-    "HRSCD", "InriaAIL", "Tiselac", "GID15", "ZueriCrop", "AID", "DubaiSegmentation"
+    "HRSCD", "InriaAIL", "Tiselac", "GID15", "ZueriCrop", "AID", "DubaiSegmentation",
+    "HKHGlacierMapping"
 ]

torchrs/datasets/dubai_segmentation.py

@@ -10,9 +10,8 @@
 
 
 class DubaiSegmentation(torch.utils.data.Dataset):
-    """ Inria Aerial Image Labeling dataset from 'Can semantic labeling methods
-    generalize to any city? the inria aerial image labeling benchmark', Maggiori et al. (2017)
-    https://ieeexplore.ieee.org/document/8127684
+    """ Semantic segmentation dataset of Dubai imagery taken by MBRSC satellites
+    https://humansintheloop.org/resources/datasets/semantic-segmentation-dataset/
 
     """
     classes = {

torchrs/datasets/hkh_glacier.py

@@ -0,0 +1,61 @@
+import os
+from glob import glob
+from typing import List, Dict
+
+import torch
+import numpy as np
+
+from torchrs.transforms import Compose, ToTensor
+
+
+class HKHGlacierMapping(torch.utils.data.Dataset):
+    """ Hindu Kush Himalayas (HKH) Glacier Mapping dataset
+    https://lila.science/datasets/hkh-glacier-mapping
+
+    'We also provide 14190 numpy patches. The numpy patches are all of size 512x512x15 and
+    corresponding 512x512x2 pixel-wise mask labels; the two channels in the pixel-wise masks
+    correspond to clean-iced and debris-covered glaciers. Patches' geolocation information,
+    time stamps, source Landsat IDs, and glacier density are available in a geojson metadata file.'
+
+    """
+    bands = [
+        "LE7 B1 (blue)",
+        "LE7 B2 (green)",
+        "LE7 B3 (red)",
+        "LE7 B4 (near infrared)",
+        "LE7 B5 (shortwave infrared 1)",
+        "LE7 B6_VCID_1 (low-gain thermal infrared)",
+        "LE7 B6_VCID_2 (high-gain thermal infrared)",
+        "LE7 B7 (shortwave infrared 2)",
+        "LE7 B8 (panchromatic)",
+        "LE7 BQA (quality bitmask)",
+        "NDVI (vegetation index)",
+        "NDSI (snow index)",
+        "NDWI (water index)",
+        "SRTM 90 elevation",
+        "SRTM 90 slope"
+    ]
+
+    def __init__(
+        self,
+        root: str = ".data/hkh_glacier_mapping",
+        transform: Compose = Compose([ToTensor()]),
+    ):
+        self.transform = transform
+        self.images = self.load_images(root)
+
+    @staticmethod
+    def load_images(path: str) -> List[Dict]:
+        images = sorted(glob(os.path.join(path, "images", "*.npy")))
+        masks = sorted(glob(os.path.join(path, "masks", "*.npy")))
+        return [dict(image=image, mask=mask) for image, mask in zip(images, masks)]
+
+    def __len__(self) -> int:
+        return len(self.images)
+
+    def __getitem__(self, idx: int) -> Dict[str, torch.Tensor]:
+        image_path, target_path = self.images[idx]["image"], self.images[idx]["mask"]
+        x, y = np.load(image_path), np.load(target_path)
+        y0, y1 = y[..., 0], y[..., 1]
+        x, y0, y1 = self.transform([x, y0, y1])
+        return dict(x=x, clean_ice_mask=y0, debris_covered_mask=y1)

torchrs/train/datamodules/__init__.py

@@ -21,6 +21,7 @@
 from .zuericrop import ZueriCropDataModule
 from .aid import AIDDataModule
 from .dubai_segmentation import DubaiSegmentationDataModule
+from .hkh_glacier import HKHGlacierMappingDataModule
 
 
 __all__ = [
@@ -29,5 +30,6 @@
     "RSICDDataModule", "OSCDDataModule", "S2LookingDataModule", "LEVIRCDPlusDataModule",
     "FAIR1MDataModule", "SydneyCaptionsDataModule", "UCMCaptionsDataModule", "S2MTCPDataModule",
     "ADVANCEDataModule", "SAT4DataModule", "SAT6DataModule", "HRSCDDataModule", "InriaAILDataModule",
-    "TiselacDataModule", "GID15DataModule", "ZueriCropDataModule", "AIDDataModule", "DubaiSegmentationDataModule"
+    "TiselacDataModule", "GID15DataModule", "ZueriCropDataModule", "AIDDataModule",
+    "DubaiSegmentationDataModule", "HKHGlacierMappingDataModule"
 ]

torchrs/train/datamodules/dubai_segmentation.py

@@ -23,4 +23,3 @@ def setup(self, stage: Optional[str] = None):
         self.train_dataset, self.val_dataset, self.test_dataset = dataset_split(
             dataset, val_pct=self.val_split, test_pct=self.test_split
         )
-

torchrs/train/datamodules/hkh_glacier.py

@@ -0,0 +1,25 @@
+from typing import Optional
+
+from torchrs.transforms import Compose, ToTensor
+from torchrs.datasets.utils import dataset_split
+from torchrs.train.datamodules import BaseDataModule
+from torchrs.datasets import HKHGlacierMapping
+
+
+class HKHGlacierMappingDataModule(BaseDataModule):
+
+    def __init__(
+        self,
+        root: str = ".data/hkh_glacier_mapping",
+        transform: Compose = Compose([ToTensor()]),
+        *args, **kwargs
+    ):
+        super().__init__(*args, **kwargs)
+        self.root = root
+        self.transform = transform
+
+    def setup(self, stage: Optional[str] = None):
+        dataset = HKHGlacierMapping(root=self.root, transform=self.transform)
+        self.train_dataset, self.val_dataset, self.test_dataset = dataset_split(
+            dataset, val_pct=self.val_split, test_pct=self.test_split
+        )