Move conf files to experiment subdirs (#1660)

conf/bigearthnet.yaml

@@ -1,22 +0,0 @@
-trainer:
-  min_epochs: 15
-  max_epochs: 40
-model:
-  class_path: MultiLabelClassificationTask
-  init_args:
-    loss: "bce"
-    model: "resnet18"
-    lr: 1e-3
-    patience: 6
-    weights: null
-    in_channels: 14
-    num_classes: 19
-data:
-  class_path: BigEarthNetDataModule
-  init_args:
-    batch_size: 128
-    num_workers: 4
-  dict_kwargs:
-    root: "data/bigearthnet"
-    bands: "all"
-    num_classes: ${model.init_args.num_classes}

conf/cyclone.yaml

@@ -1,19 +0,0 @@
-trainer:
-  min_epochs: 15
-  max_epochs: 40
-model:
-  class_path: RegressionTask
-  init_args:
-    model: "resnet18"
-    weights: null
-    num_outputs: 1
-    in_channels: 3
-    lr: 1e-3
-    patience: 2
-data:
-  class_path: TropicalCycloneDataModule
-  init_args:
-    batch_size: 32
-    num_workers: 4
-  dict_kwargs:
-    root: "data/cyclone"

conf/deepglobelandcover.yaml

@@ -1,25 +0,0 @@
-trainer:
-  min_epochs: 15
-  max_epochs: 40
-model:
-  class_path: SemanticSegmentationTask
-  init_args:
-    loss: "ce"
-    model: "unet"
-    backbone: "resnet18"
-    weights: null
-    lr: 1e-3
-    patience: 6
-    in_channels: 3
-    num_classes: 7
-    num_filters: 1
-    ignore_index: null
-data:
-  class_path: DeepGlobeLandCoverDataModule
-  init_args:
-    batch_size: 1
-    patch_size: 64
-    val_split_pct: 0.5
-    num_workers: 0
-  dict_kwargs:
-    root: "data/deepglobelandcover"

conf/gid15.yaml

@@ -1,25 +0,0 @@
-trainer:
-  min_epochs: 15
-  max_epochs: 40
-model:
-  class_path: SemanticSegmentationTask
-  init_args:
-    loss: "ce"
-    model: "unet"
-    backbone: "resnet18"
-    weights: null
-    lr: 1e-3
-    patience: 6
-    in_channels: 3
-    num_classes: 16
-    num_filters: 1
-    ignore_index: null
-data:
-  class_path: GID15DataModule
-  init_args:
-    batch_size: 1
-    patch_size: 64
-    val_split_pct: 0.5
-    num_workers: 0
-  dict_kwargs:
-    root: "data/gid15"

conf/inria.yaml

@@ -1,23 +0,0 @@
-trainer:
-  min_epochs: 15
-  max_epochs: 40
-model:
-  class_path: SemanticSegmentationTask
-  init_args:
-    loss: "ce"
-    model: "unet"
-    backbone: "resnet18"
-    weights: true
-    lr: 1e-3
-    patience: 6
-    in_channels: 3
-    num_classes: 2
-    ignore_index: null
-data:
-  class_path: InriaAerialImageLabelingDataModule
-  init_args:
-    batch_size: 1
-    patch_size: 512
-    num_workers: 32
-  dict_kwargs:
-    root: "data/inria"

conf/naipchesapeake.yaml

@@ -1,25 +0,0 @@
-trainer:
-  min_epochs: 15
-  max_epochs: 40
-model:
-  class_path: SemanticSegmentationTask
-  init_args:
-    loss: "ce"
-    model: "deeplabv3+"
-    backbone: "resnet34"
-    weights: true
-    lr: 1e-3
-    patience: 2
-    in_channels: 4
-    num_classes: 14
-    num_filters: 64
-    ignore_index: null
-data:
-  class_path: NAIPChesapeakeDataModule
-  init_args:
-    batch_size: 32
-    num_workers: 4
-    patch_size: 32
-  dict_kwargs:
-    naip_paths: "data/naip"
-    chesapeake_paths: "data/chesapeake/BAYWIDE"

conf/nasa_marine_debris.yaml

@@ -1,19 +0,0 @@
-trainer:
-  min_epochs: 15
-  max_epochs: 40
-model:
-  class_path: ObjectDetectionTask
-  init_args:
-    model: "faster-rcnn"
-    backbone: "resnet50"
-    num_classes: 2
-    lr: 1.2e-4
-    patience: 6
-data:
-  class_path: NASAMarineDebrisDataModule
-  init_args:
-    batch_size: 4
-    num_workers: 6
-    val_split_pct: 0.2
-  dict_kwargs:
-    root: "data/nasamr/nasa_marine_debris"

conf/potsdam2d.yaml

@@ -1,25 +0,0 @@
-trainer:
-  min_epochs: 15
-  max_epochs: 40
-model:
-  class_path: SemanticSegmentationTask
-  init_args:
-    loss: "ce"
-    model: "unet"
-    backbone: "resnet18"
-    weights: null
-    lr: 1e-3
-    patience: 6
-    in_channels: 4
-    num_classes: 6
-    num_filters: 1
-    ignore_index: null
-data:
-  class_path: Potsdam2DDataModule
-  init_args:
-    batch_size: 1
-    patch_size: 64
-    val_split_pct: 0.5
-    num_workers: 0
-  dict_kwargs:
-    root: "data/potsdam"

conf/seco_100k.yaml

@@ -1,22 +0,0 @@
-trainer:
-  min_epochs: 15
-  max_epochs: 40
-model:
-  class_path: BYOLTask
-  init_args:
-    in_channels: 12
-    backbone: "resnet18"
-    weights: True
-    lr: 1e-3
-    patience: 6
-    optimizer: "Adam"
-data:
-  class_path: SeasonalContrastS2DataModule
-  init_args:
-    batch_size: 64
-    num_workers: 16
-  dict_kwargs:
-    root: "data/seco"
-    version: "100k"
-    seasons: 2
-    bands: ["B1", "B2", "B3", "B4", "B5", "B6", "B7", "B8", "B8A", "B9", "B11", "B12"]

conf/sen12ms.yaml

@@ -1,23 +0,0 @@
-trainer:
-  min_epochs: 15
-  max_epochs: 40
-model:
-  class_path: SemanticSegmentationTask
-  init_args:
-    loss: "ce"
-    model: "unet"
-    backbone: "resnet18"
-    weights: null
-    lr: 1e-3
-    patience: 2
-    in_channels: 15
-    num_classes: 11
-    ignore_index: null
-data:
-  class_path: SEN12MSDataModule
-  init_args:
-    batch_size: 32
-    num_workers: 4
-  dict_kwargs:
-    root: "data/sen12ms"
-    band_set: "all"

conf/spacenet1.yaml

@@ -1,22 +0,0 @@
-trainer:
-  min_epochs: 15
-  max_epochs: 40
-model:
-  class_path: SemanticSegmentationTask
-  init_args:
-    loss: "ce"
-    model: "unet"
-    backbone: "resnet18"
-    weights: true
-    lr: 1e-3
-    patience: 6
-    in_channels: 3
-    num_classes: 3
-    ignore_index: 0
-data:
-  class_path: SpaceNet1DataModule
-  init_args:
-    batch_size: 32
-    num_workers: 4
-  dict_kwargs:
-    root: "data/spacenet"

conf/vaihingen2d.yaml

@@ -1,25 +0,0 @@
-trainer:
-  min_epochs: 15
-  max_epochs: 40
-model:
-  class_path: SemanticSegmentationTask
-  init_args:
-    loss: "ce"
-    model: "unet"
-    backbone: "resnet18"
-    weights: null
-    lr: 1e-3
-    patience: 6
-    in_channels: 3
-    num_classes: 7
-    num_filters: 1
-    ignore_index: null
-data:
-  class_path: Vaihingen2DDataModule
-  init_args:
-    batch_size: 1
-    patch_size: 64
-    val_split_pct: 0.5
-    num_workers: 0
-  dict_kwargs:
-    root: "data/vaihingen"

experiments/ssl4eo/landsat/README.md

@@ -89,10 +89,10 @@ This will create patches of NLCD and CDL data with the same locations and dimens
 Using either the newly created datasets or after downloading the datasets from Hugging Face, you can run each experiment using:
 
 ```console
-$ python3 ../../../train.py config_file=...
+$ torchgeo --config *.yaml
 ```
 
-The config files to be passed can be found in the `../../../conf/` directory. Feel free to tweak any hyperparameters you see in these files. The default values are the optimal hyperparameters we found.
+The config files to be passed can be found in the `conf/` directory. Feel free to tweak any hyperparameters you see in these files. The default values are the optimal hyperparameters we found.
 
 ## Plotting