data_loader_map_provider

Summary: replace dataloader_zoo with a pluggable DataLoaderMapProvider.

Reviewed By: shapovalov

Differential Revision: D36475441

fbshipit-source-id: d16abb190d876940434329928f2e3f2794a25416

projects/implicitron_trainer/README.md

@@ -237,7 +237,7 @@ generic_model_args: GenericModel
 solver_args: init_optimizer
 data_source_args: ImplicitronDataSource
 └-- dataset_map_provider_*_args
-└-- dataloader_args
+└-- data_loader_map_provider_*_args
 ```
 
 Please look at the annotations of the respective classes or functions for the lists of hyperparameters.

projects/implicitron_trainer/configs/repro_base.yaml

@@ -7,7 +7,7 @@ visualize_interval: 0
 visdom_port: 8097
 data_source_args:
   dataset_provider_class_type: JsonIndexDatasetMapProvider
-  dataloader_args:
+  data_loader_map_provider_SequenceDataLoaderMapProvider_args:
     batch_size: 10
     dataset_len: 1000
     dataset_len_val: 1

projects/implicitron_trainer/configs/repro_multiseq_base.yaml

@@ -2,7 +2,7 @@ defaults:
 - repro_base.yaml
 - _self_
 data_source_args:
-  dataloader_args:
+  data_loader_map_provider_SequenceDataLoaderMapProvider_args:
     batch_size: 10
     dataset_len: 1000
     dataset_len_val: 1

projects/implicitron_trainer/configs/repro_singleseq_base.yaml

@@ -2,7 +2,7 @@ defaults:
 - repro_base
 - _self_
 data_source_args:
-  dataloader_args:
+  data_loader_map_provider_SequenceDataLoaderMapProvider_args:
     batch_size: 1
     dataset_len: 1000
     dataset_len_val: 1

projects/implicitron_trainer/configs/repro_singleseq_wce_base.yaml

@@ -2,7 +2,7 @@ defaults:
 - repro_singleseq_base
 - _self_
 data_source_args:
-  dataloader_args:
+  data_loader_map_provider_SequenceDataLoaderMapProvider_args:
     batch_size: 10
     dataset_len: 1000
     dataset_len_val: 1

projects/implicitron_trainer/experiment.py

@@ -64,8 +64,8 @@ import tqdm
 from omegaconf import DictConfig, OmegaConf
 from packaging import version
 from pytorch3d.implicitron.dataset import utils as ds_utils
+from pytorch3d.implicitron.dataset.data_loader_map_provider import DataLoaderMap
 from pytorch3d.implicitron.dataset.data_source import ImplicitronDataSource, Task
-from pytorch3d.implicitron.dataset.dataloader_zoo import Dataloaders
 from pytorch3d.implicitron.dataset.dataset_base import FrameData
 from pytorch3d.implicitron.dataset.dataset_map_provider import DatasetMap
 from pytorch3d.implicitron.dataset.implicitron_dataset import ImplicitronDataset
@@ -553,7 +553,7 @@ def _eval_and_dump(
     cfg,
     task: Task,
     datasets: DatasetMap,
-    dataloaders: Dataloaders,
+    dataloaders: DataLoaderMap,
     model,
     stats,
     device,
@@ -566,7 +566,7 @@ def _eval_and_dump(
     dataloader = dataloaders.test
 
     if dataloader is None:
-        raise ValueError('Dataloaders have to contain the "test" entry for eval!')
+        raise ValueError('DataLoaderMap have to contain the "test" entry for eval!')
 
     if task == Task.SINGLE_SEQUENCE:
         if datasets.train is None:

pytorch3d/implicitron/dataset/data_loader_map_provider.py

@@ -0,0 +1,139 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+from dataclasses import dataclass
+from typing import Optional, Sequence
+
+import torch
+from pytorch3d.implicitron.tools.config import registry, ReplaceableBase
+
+from .dataset_base import FrameData, ImplicitronDatasetBase
+from .dataset_map_provider import DatasetMap
+from .scene_batch_sampler import SceneBatchSampler
+
+
+@dataclass
+class DataLoaderMap:
+    """
+    A collection of data loaders for Implicitron.
+
+    Members:
+
+        train: a data loader for training
+        val: a data loader for validating during training
+        test: a data loader for final evaluation
+    """
+
+    train: Optional[torch.utils.data.DataLoader[FrameData]]
+    val: Optional[torch.utils.data.DataLoader[FrameData]]
+    test: Optional[torch.utils.data.DataLoader[FrameData]]
+
+    def __getitem__(
+        self, split: str
+    ) -> Optional[torch.utils.data.DataLoader[FrameData]]:
+        """
+        Get one of the data loaders by key (name of data split)
+        """
+        if split not in ["train", "val", "test"]:
+            raise ValueError(f"{split} was not a valid split name (train/val/test)")
+        return getattr(self, split)
+
+
+class DataLoaderMapProviderBase(ReplaceableBase):
+    """
+    Provider of a collection of data loaders for a given collection of datasets.
+    """
+
+    def get_data_loader_map(self, datasets: DatasetMap) -> DataLoaderMap:
+        """
+        Returns a collection of data loaders for a given collection of datasets.
+        """
+        raise NotImplementedError()
+
+
+@registry.register
+class SequenceDataLoaderMapProvider(DataLoaderMapProviderBase):
+    """
+    The default implementation of DataLoaderMapProviderBase.
+
+    Members:
+        batch_size: The size of the batch of the data loader.
+        num_workers: Number data-loading threads.
+        dataset_len: The number of batches in a training epoch.
+        dataset_len_val: The number of batches in a validation epoch.
+        images_per_seq_options: Possible numbers of images sampled per sequence.
+        sample_consecutive_frames: if True, will sample a contiguous interval of frames
+            in the sequence. It first sorts the frames by timestimps when available,
+            otherwise by frame numbers, finds the connected segments within the sequence
+            of sufficient length, then samples a random pivot element among them and
+            ideally uses it as a middle of the temporal window, shifting the borders
+            where necessary. This strategy mitigates the bias against shorter segments
+            and their boundaries.
+        consecutive_frames_max_gap: if a number > 0, then used to define the maximum
+            difference in frame_number of neighbouring frames when forming connected
+            segments; if both this and consecutive_frames_max_gap_seconds are 0s,
+            the whole sequence is considered a segment regardless of frame numbers.
+        consecutive_frames_max_gap_seconds: if a number > 0.0, then used to define the
+            maximum difference in frame_timestamp of neighbouring frames when forming
+            connected segments; if both this and consecutive_frames_max_gap are 0s,
+            the whole sequence is considered a segment regardless of frame timestamps.
+    """
+
+    batch_size: int = 1
+    num_workers: int = 0
+    dataset_len: int = 1000
+    dataset_len_val: int = 1
+    images_per_seq_options: Sequence[int] = (2,)
+    sample_consecutive_frames: bool = False
+    consecutive_frames_max_gap: int = 0
+    consecutive_frames_max_gap_seconds: float = 0.1
+
+    def get_data_loader_map(self, datasets: DatasetMap) -> DataLoaderMap:
+        """
+        Returns a collection of data loaders for a given collection of datasets.
+        """
+
+        data_loader_kwargs = {
+            "num_workers": self.num_workers,
+            "collate_fn": FrameData.collate,
+        }
+
+        def train_or_val_loader(
+            dataset: Optional[ImplicitronDatasetBase], num_batches: int
+        ) -> Optional[torch.utils.data.DataLoader]:
+            if dataset is None:
+                return None
+            batch_sampler = SceneBatchSampler(
+                dataset,
+                self.batch_size,
+                num_batches=len(dataset) if num_batches <= 0 else num_batches,
+                images_per_seq_options=self.images_per_seq_options,
+                sample_consecutive_frames=self.sample_consecutive_frames,
+                consecutive_frames_max_gap=self.consecutive_frames_max_gap,
+                consecutive_frames_max_gap_seconds=self.consecutive_frames_max_gap_seconds,
+            )
+            return torch.utils.data.DataLoader(
+                dataset,
+                batch_sampler=batch_sampler,
+                **data_loader_kwargs,
+            )
+
+        train_data_loader = train_or_val_loader(datasets.train, self.dataset_len)
+        val_data_loader = train_or_val_loader(datasets.val, self.dataset_len_val)
+
+        test_dataset = datasets.test
+        if test_dataset is not None:
+            test_data_loader = torch.utils.data.DataLoader(
+                test_dataset,
+                batch_sampler=test_dataset.get_eval_batches(),
+                **data_loader_kwargs,
+            )
+        else:
+            test_data_loader = None
+
+        return DataLoaderMap(
+            train=train_data_loader, val=val_data_loader, test=test_data_loader
+        )

pytorch3d/implicitron/dataset/data_source.py

@@ -6,15 +6,10 @@
 
 from typing import Tuple
 
-from omegaconf import DictConfig
+from pytorch3d.implicitron.tools.config import ReplaceableBase, run_auto_creation
-from pytorch3d.implicitron.tools.config import (
-    get_default_args_field,
-    ReplaceableBase,
-    run_auto_creation,
-)
 
 from . import json_index_dataset_map_provider  # noqa
-from .dataloader_zoo import dataloader_zoo, Dataloaders
+from .data_loader_map_provider import DataLoaderMap, DataLoaderMapProviderBase
 from .dataset_map_provider import DatasetMap, DatasetMapProviderBase, Task
 
 
@@ -24,7 +19,7 @@ class DataSourceBase(ReplaceableBase):
     and DataLoader configuration.
     """
 
-    def get_datasets_and_dataloaders(self) -> Tuple[DatasetMap, Dataloaders]:
+    def get_datasets_and_dataloaders(self) -> Tuple[DatasetMap, DataLoaderMap]:
         raise NotImplementedError()
 
 
@@ -36,18 +31,20 @@ class ImplicitronDataSource(DataSourceBase):  # pyre-ignore[13]
     Members:
         dataset_map_provider_class_type: identifies type for dataset_map_provider.
             e.g. JsonIndexDatasetMapProvider for Co3D.
+        data_loader_map_provider_class_type: identifies type for data_loader_map_provider.
     """
 
     dataset_map_provider: DatasetMapProviderBase
     dataset_map_provider_class_type: str
-    dataloader_args: DictConfig = get_default_args_field(dataloader_zoo)
+    data_loader_map_provider: DataLoaderMapProviderBase
+    data_loader_map_provider_class_type: str = "SequenceDataLoaderMapProvider"
 
     def __post_init__(self):
         run_auto_creation(self)
 
-    def get_datasets_and_dataloaders(self) -> Tuple[DatasetMap, Dataloaders]:
+    def get_datasets_and_dataloaders(self) -> Tuple[DatasetMap, DataLoaderMap]:
         datasets = self.dataset_map_provider.get_dataset_map()
-        dataloaders = dataloader_zoo(datasets, **self.dataloader_args)
+        dataloaders = self.data_loader_map_provider.get_data_loader_map(datasets)
         return datasets, dataloaders
 
     def get_task(self) -> Task:

pytorch3d/implicitron/dataset/dataloader_zoo.py

@@ -1,116 +0,0 @@
-# Copyright (c) Meta Platforms, Inc. and affiliates.
-# All rights reserved.
-#
-# This source code is licensed under the BSD-style license found in the
-# LICENSE file in the root directory of this source tree.
-
-from dataclasses import dataclass
-from typing import Optional, Sequence
-
-import torch
-from pytorch3d.implicitron.tools.config import enable_get_default_args
-
-from .dataset_base import FrameData, ImplicitronDatasetBase
-from .dataset_map_provider import DatasetMap
-from .scene_batch_sampler import SceneBatchSampler
-
-
-@dataclass
-class Dataloaders:
-    """
-    A provider of dataloaders for implicitron.
-
-    Members:
-
-        train: a dataloader for training
-        val: a dataloader for validating during training
-        test: a dataloader for final evaluation
-    """
-
-    train: Optional[torch.utils.data.DataLoader[FrameData]]
-    val: Optional[torch.utils.data.DataLoader[FrameData]]
-    test: Optional[torch.utils.data.DataLoader[FrameData]]
-
-
-def dataloader_zoo(
-    datasets: DatasetMap,
-    batch_size: int = 1,
-    num_workers: int = 0,
-    dataset_len: int = 1000,
-    dataset_len_val: int = 1,
-    images_per_seq_options: Sequence[int] = (2,),
-    sample_consecutive_frames: bool = False,
-    consecutive_frames_max_gap: int = 0,
-    consecutive_frames_max_gap_seconds: float = 0.1,
-) -> Dataloaders:
-    """
-    Returns a set of dataloaders for a given set of datasets.
-
-    Args:
-        datasets: A dictionary containing the
-            `"dataset_subset_name": torch_dataset_object` key, value pairs.
-        batch_size: The size of the batch of the dataloader.
-        num_workers: Number data-loading threads.
-        dataset_len: The number of batches in a training epoch.
-        dataset_len_val: The number of batches in a validation epoch.
-        images_per_seq_options: Possible numbers of images sampled per sequence.
-        sample_consecutive_frames: if True, will sample a contiguous interval of frames
-            in the sequence. It first sorts the frames by timestimps when available,
-            otherwise by frame numbers, finds the connected segments within the sequence
-            of sufficient length, then samples a random pivot element among them and
-            ideally uses it as a middle of the temporal window, shifting the borders
-            where necessary. This strategy mitigates the bias against shorter segments
-            and their boundaries.
-        consecutive_frames_max_gap: if a number > 0, then used to define the maximum
-            difference in frame_number of neighbouring frames when forming connected
-            segments; if both this and consecutive_frames_max_gap_seconds are 0s,
-            the whole sequence is considered a segment regardless of frame numbers.
-        consecutive_frames_max_gap_seconds: if a number > 0.0, then used to define the
-            maximum difference in frame_timestamp of neighbouring frames when forming
-            connected segments; if both this and consecutive_frames_max_gap are 0s,
-            the whole sequence is considered a segment regardless of frame timestamps.
-
-    Returns:
-        dataloaders: A dictionary containing the
-            `"dataset_subset_name": torch_dataloader_object` key, value pairs.
-    """
-
-    dataloader_kwargs = {"num_workers": num_workers, "collate_fn": FrameData.collate}
-
-    def train_or_val_loader(
-        dataset: Optional[ImplicitronDatasetBase], num_batches: int
-    ) -> Optional[torch.utils.data.DataLoader]:
-        if dataset is None:
-            return None
-        batch_sampler = SceneBatchSampler(
-            dataset,
-            batch_size,
-            num_batches=len(dataset) if num_batches <= 0 else num_batches,
-            images_per_seq_options=images_per_seq_options,
-            sample_consecutive_frames=sample_consecutive_frames,
-            consecutive_frames_max_gap=consecutive_frames_max_gap,
-            consecutive_frames_max_gap_seconds=consecutive_frames_max_gap_seconds,
-        )
-        return torch.utils.data.DataLoader(
-            dataset,
-            batch_sampler=batch_sampler,
-            **dataloader_kwargs,
-        )
-
-    train_dataloader = train_or_val_loader(datasets.train, dataset_len)
-    val_dataloader = train_or_val_loader(datasets.val, dataset_len_val)
-
-    test_dataset = datasets.test
-    if test_dataset is not None:
-        test_dataloader = torch.utils.data.DataLoader(
-            test_dataset,
-            batch_sampler=test_dataset.get_eval_batches(),
-            **dataloader_kwargs,
-        )
-    else:
-        test_dataloader = None
-
-    return Dataloaders(train=train_dataloader, val=val_dataloader, test=test_dataloader)
-
-
-enable_get_default_args(dataloader_zoo)

tests/implicitron/data/data_source.yaml

@@ -1,14 +1,5 @@
 dataset_map_provider_class_type: ???
-dataloader_args:
+data_loader_map_provider_class_type: SequenceDataLoaderMapProvider
-  batch_size: 1
-  num_workers: 0
-  dataset_len: 1000
-  dataset_len_val: 1
-  images_per_seq_options:
-  - 2
-  sample_consecutive_frames: false
-  consecutive_frames_max_gap: 0
-  consecutive_frames_max_gap_seconds: 0.1
 dataset_map_provider_JsonIndexDatasetMapProvider_args:
   category: ???
   task_str: singlesequence
@@ -31,3 +22,13 @@ dataset_map_provider_JsonIndexDatasetMapProvider_args:
     image_height: 800
     remove_empty_masks: true
   path_manager: null
+data_loader_map_provider_SequenceDataLoaderMapProvider_args:
+  batch_size: 1
+  num_workers: 0
+  dataset_len: 1000
+  dataset_len_val: 1
+  images_per_seq_options:
+  - 2
+  sample_consecutive_frames: false
+  consecutive_frames_max_gap: 0
+  consecutive_frames_max_gap_seconds: 0.1