CO3Dv2 multi-category extension

Summary:
Allows loading of multiple categories.
Multiple categories are provided in a comma-separated list of category names.

Reviewed By: bottler, shapovalov

Differential Revision: D40803297

fbshipit-source-id: 863938be3aa6ffefe9e563aede4a2e9e66aeeaa8

projects/implicitron_trainer/tests/experiment.yaml

@@ -62,6 +62,7 @@ data_source_ImplicitronDataSource_args:
     test_on_train: false
     only_test_set: false
     load_eval_batches: true
+    num_load_workers: 4
     n_known_frames_for_test: 0
     dataset_class_type: JsonIndexDataset
     path_manager_factory_class_type: PathManagerFactory

pytorch3d/implicitron/dataset/dataset_base.py

@@ -9,6 +9,7 @@ from dataclasses import dataclass, field, fields
 from typing import (
     Any,
     ClassVar,
+    Dict,
     Iterable,
     Iterator,
     List,
@@ -259,6 +260,12 @@ class DatasetBase(_GenericWorkaround, torch.utils.data.Dataset[FrameData]):
         """
         raise ValueError("This dataset does not contain videos.")
 
+    def join(self, other_datasets: Iterable["DatasetBase"]) -> None:
+        """
+        Joins the current dataset with a list of other datasets of the same type.
+        """
+        raise NotImplementedError()
+
     def get_eval_batches(self) -> Optional[List[List[int]]]:
         return None
 
@@ -267,6 +274,22 @@ class DatasetBase(_GenericWorkaround, torch.utils.data.Dataset[FrameData]):
         # pyre-ignore[16]
         return self._seq_to_idx.keys()
 
+    def category_to_sequence_names(self) -> Dict[str, List[str]]:
+        """
+        Returns a dict mapping from each dataset category to a list of its
+        sequence names.
+
+        Returns:
+            category_to_sequence_names: Dict {category_i: [..., sequence_name_j, ...]}
+        """
+        c2seq = defaultdict(list)
+        for sequence_name in self.sequence_names():
+            first_frame_idx = next(self.sequence_indices_in_order(sequence_name))
+            # crashes without overriding __getitem__
+            sequence_category = self[first_frame_idx].sequence_category
+            c2seq[sequence_category].append(sequence_name)
+        return dict(c2seq)
+
     def sequence_frames_in_order(
         self, seq_name: str
     ) -> Iterator[Tuple[float, int, int]]:

pytorch3d/implicitron/dataset/dataset_map_provider.py

@@ -7,7 +7,7 @@
 import logging
 import os
 from dataclasses import dataclass
-from typing import Iterator, Optional
+from typing import Iterable, Iterator, Optional
 
 from iopath.common.file_io import PathManager
 from pytorch3d.implicitron.tools.config import registry, ReplaceableBase
@@ -51,6 +51,34 @@ class DatasetMap:
         if self.test is not None:
             yield self.test
 
+    def join(self, other_dataset_maps: Iterable["DatasetMap"]) -> None:
+        """
+        Joins the current DatasetMap with other dataset maps from the input list.
+
+        For each subset of each dataset map (train/val/test), the function
+        omits joining the subsets that are None.
+
+        Note the train/val/test datasets of the current dataset map will be
+        modified in-place.
+
+        Args:
+            other_dataset_maps: The list of dataset maps to be joined into the
+                current dataset map.
+        """
+        for set_ in ["train", "val", "test"]:
+            dataset_list = [
+                getattr(self, set_),
+                *[getattr(dmap, set_) for dmap in other_dataset_maps],
+            ]
+            dataset_list = [d for d in dataset_list if d is not None]
+            if len(dataset_list) == 0:
+                setattr(self, set_, None)
+                continue
+            d0 = dataset_list[0]
+            if len(dataset_list) > 1:
+                d0.join(dataset_list[1:])
+            setattr(self, set_, d0)
+
 
 class DatasetMapProviderBase(ReplaceableBase):
     """

pytorch3d/implicitron/dataset/json_index_dataset.py

@@ -19,6 +19,8 @@ from pathlib import Path
 from typing import (
     Any,
     ClassVar,
+    Dict,
+    Iterable,
     List,
     Optional,
     Sequence,
@@ -188,7 +190,44 @@ class JsonIndexDataset(DatasetBase, ReplaceableBase):
                 self.eval_batch_index
             )
 
-    def is_filtered(self):
+    def join(self, other_datasets: Iterable[DatasetBase]) -> None:
+        """
+        Join the dataset with other JsonIndexDataset objects.
+
+        Args:
+            other_datasets: A list of JsonIndexDataset objects to be joined
+                into the current dataset.
+        """
+        if not all(isinstance(d, JsonIndexDataset) for d in other_datasets):
+            raise ValueError("This function can only join a list of JsonIndexDataset")
+        # pyre-ignore[16]
+        self.frame_annots.extend([fa for d in other_datasets for fa in d.frame_annots])
+        # pyre-ignore[16]
+        self.seq_annots.update(
+            # https://gist.github.com/treyhunner/f35292e676efa0be1728
+            functools.reduce(
+                lambda a, b: {**a, **b},
+                [d.seq_annots for d in other_datasets],  # pyre-ignore[16]
+            )
+        )
+        all_eval_batches = [
+            self.eval_batches,
+            # pyre-ignore
+            *[d.eval_batches for d in other_datasets],
+        ]
+        if not (
+            all(ba is None for ba in all_eval_batches)
+            or all(ba is not None for ba in all_eval_batches)
+        ):
+            raise ValueError(
+                "When joining datasets, either all joined datasets have to have their"
+                " eval_batches defined, or all should have their eval batches undefined."
+            )
+        if self.eval_batches is not None:
+            self.eval_batches = sum(all_eval_batches, [])
+        self._invalidate_indexes(filter_seq_annots=True)
+
+    def is_filtered(self) -> bool:
         """
         Returns `True` in case the dataset has been filtered and thus some frame annotations
         stored on the disk might be missing in the dataset object.
@@ -211,6 +250,7 @@ class JsonIndexDataset(DatasetBase, ReplaceableBase):
         seq_frame_index: List[List[Union[Tuple[str, int, str], Tuple[str, int]]]],
         allow_missing_indices: bool = False,
         remove_missing_indices: bool = False,
+        suppress_missing_index_warning: bool = True,
     ) -> List[List[Union[Optional[int], int]]]:
         """
         Obtain indices into the dataset object given a list of frame ids.
@@ -228,6 +268,11 @@ class JsonIndexDataset(DatasetBase, ReplaceableBase):
                 If `False`, returns `None` in place of `seq_frame_index` entries that
                 are not present in the dataset.
                 If `True` removes missing indices from the returned indices.
+            suppress_missing_index_warning:
+                Active if `allow_missing_indices==True`. Suppressess a warning message
+                in case an entry from `seq_frame_index` is missing in the dataset
+                (expected in certain cases - e.g. when setting
+                `self.remove_empty_masks=True`).
 
         Returns:
             dataset_idx: Indices of dataset entries corresponding to`seq_frame_index`.
@@ -254,7 +299,8 @@ class JsonIndexDataset(DatasetBase, ReplaceableBase):
                 )
                 if not allow_missing_indices:
                     raise IndexError(msg)
-                warnings.warn(msg)
+                if not suppress_missing_index_warning:
+                    warnings.warn(msg)
                 return idx
             if path is not None:
                 # Check that the loaded frame path is consistent
@@ -288,6 +334,21 @@ class JsonIndexDataset(DatasetBase, ReplaceableBase):
         frame_index: List[Union[Tuple[str, int], Tuple[str, int, str]]],
         allow_missing_indices: bool = True,
     ) -> "JsonIndexDataset":
+        """
+        Generate a dataset subset given the list of frames specified in `frame_index`.
+
+        Args:
+            frame_index: The list of frame indentifiers (as stored in the metadata)
+                specified as `List[Tuple[sequence_name:str, frame_number:int]]`. Optionally,
+                Image paths relative to the dataset_root can be stored specified as well:
+                `List[Tuple[sequence_name:str, frame_number:int, image_path:str]]`,
+                in the latter case, if imaga_path do not match the stored paths, an error
+                is raised.
+            allow_missing_indices: If `False`, throws an IndexError upon reaching the first
+                entry from `frame_index` which is missing in the dataset.
+                Otherwise, generates a subset consisting of frames entries that actually
+                exist in the dataset.
+        """
         # Get the indices into the frame annots.
         dataset_indices = self.seq_frame_index_to_dataset_index(
             [frame_index],
@@ -838,6 +899,13 @@ class JsonIndexDataset(DatasetBase, ReplaceableBase):
             )
         return out
 
+    def category_to_sequence_names(self) -> Dict[str, List[str]]:
+        c2seq = defaultdict(list)
+        # pyre-ignore
+        for sequence_name, sa in self.seq_annots.items():
+            c2seq[sa.category].append(sequence_name)
+        return dict(c2seq)
+
     def get_eval_batches(self) -> Optional[List[List[int]]]:
         return self.eval_batches
 

pytorch3d/implicitron/dataset/json_index_dataset_map_provider_v2.py

@@ -8,6 +8,7 @@
 import copy
 import json
 import logging
+import multiprocessing
 import os
 import warnings
 from collections import defaultdict
@@ -30,6 +31,7 @@ from pytorch3d.implicitron.tools.config import (
 )
 
 from pytorch3d.renderer.cameras import CamerasBase
+from tqdm import tqdm
 
 
 _CO3DV2_DATASET_ROOT: str = os.getenv("CO3DV2_DATASET_ROOT", "")
@@ -147,7 +149,8 @@ class JsonIndexDatasetMapProviderV2(DatasetMapProviderBase):  # pyre-ignore [13]
     (test frames can repeat across batches).
 
     Args:
-        category: The object category of the dataset.
+        category: Dataset categories to load expressed as a string of comma-separated
+            category names (e.g. `"apple,car,orange"`).
         subset_name: The name of the dataset subset. For CO3Dv2, these include
             e.g. "manyview_dev_0", "fewview_test", ...
         dataset_root: The root folder of the dataset.
@@ -173,6 +176,7 @@ class JsonIndexDatasetMapProviderV2(DatasetMapProviderBase):  # pyre-ignore [13]
     test_on_train: bool = False
     only_test_set: bool = False
     load_eval_batches: bool = True
+    num_load_workers: int = 4
 
     n_known_frames_for_test: int = 0
 
@@ -189,11 +193,33 @@ class JsonIndexDatasetMapProviderV2(DatasetMapProviderBase):  # pyre-ignore [13]
         if self.only_test_set and self.test_on_train:
             raise ValueError("Cannot have only_test_set and test_on_train")
 
-        frame_file = os.path.join(
-            self.dataset_root, self.category, "frame_annotations.jgz"
-        )
+        if "," in self.category:
+            # a comma-separated list of categories to load
+            categories = [c.strip() for c in self.category.split(",")]
+            logger.info(f"Loading a list of categories: {str(categories)}.")
+            with multiprocessing.Pool(
+                processes=min(self.num_load_workers, len(categories))
+            ) as pool:
+                category_dataset_maps = list(
+                    tqdm(
+                        pool.imap(self._load_category, categories),
+                        total=len(categories),
+                    )
+                )
+            dataset_map = category_dataset_maps[0]
+            dataset_map.join(category_dataset_maps[1:])
+
+        else:
+            # one category to load
+            dataset_map = self._load_category(self.category)
+
+        self.dataset_map = dataset_map
+
+    def _load_category(self, category: str) -> DatasetMap:
+
+        frame_file = os.path.join(self.dataset_root, category, "frame_annotations.jgz")
         sequence_file = os.path.join(
-            self.dataset_root, self.category, "sequence_annotations.jgz"
+            self.dataset_root, category, "sequence_annotations.jgz"
         )
 
         path_manager = self.path_manager_factory.get()
@@ -232,7 +258,7 @@ class JsonIndexDatasetMapProviderV2(DatasetMapProviderBase):  # pyre-ignore [13]
 
         dataset = dataset_type(**common_dataset_kwargs)
 
-        available_subset_names = self._get_available_subset_names()
+        available_subset_names = self._get_available_subset_names(category)
         logger.debug(f"Available subset names: {str(available_subset_names)}.")
         if self.subset_name not in available_subset_names:
             raise ValueError(
@@ -242,20 +268,20 @@ class JsonIndexDatasetMapProviderV2(DatasetMapProviderBase):  # pyre-ignore [13]
 
         # load the list of train/val/test frames
         subset_mapping = self._load_annotation_json(
-            os.path.join(
-                self.category, "set_lists", f"set_lists_{self.subset_name}.json"
-            )
+            os.path.join(category, "set_lists", f"set_lists_{self.subset_name}.json")
         )
 
         # load the evaluation batches
         if self.load_eval_batches:
             eval_batch_index = self._load_annotation_json(
                 os.path.join(
-                    self.category,
+                    category,
                     "eval_batches",
                     f"eval_batches_{self.subset_name}.json",
                 )
             )
+        else:
+            eval_batch_index = None
 
         train_dataset = None
         if not self.only_test_set:
@@ -313,9 +339,7 @@ class JsonIndexDatasetMapProviderV2(DatasetMapProviderBase):  # pyre-ignore [13]
                     )
                 logger.info(f"# eval batches: {len(test_dataset.eval_batches)}")
 
-        self.dataset_map = DatasetMap(
-            train=train_dataset, val=val_dataset, test=test_dataset
-        )
+        return DatasetMap(train=train_dataset, val=val_dataset, test=test_dataset)
 
     @classmethod
     def dataset_tweak_args(cls, type, args: DictConfig) -> None:
@@ -381,10 +405,10 @@ class JsonIndexDatasetMapProviderV2(DatasetMapProviderBase):  # pyre-ignore [13]
             data = json.load(f)
         return data
 
-    def _get_available_subset_names(self):
+    def _get_available_subset_names(self, category: str):
         return get_available_subset_names(
             self.dataset_root,
-            self.category,
+            category,
             path_manager=self.path_manager_factory.get(),
         )
 

pytorch3d/implicitron/dataset/scene_batch_sampler.py

@@ -6,8 +6,9 @@
 
 
 import warnings
+from collections import Counter
 from dataclasses import dataclass, field
-from typing import Iterable, Iterator, List, Sequence, Tuple
+from typing import Dict, Iterable, Iterator, List, Sequence, Tuple
 
 import numpy as np
 from torch.utils.data.sampler import Sampler
@@ -42,8 +43,17 @@ class SceneBatchSampler(Sampler[List[int]]):
     # same but for timestamps if they are available
     consecutive_frames_max_gap_seconds: float = 0.1
 
+    # if True, the sampler first reads from the dataset the mapping between
+    # sequence names and their categories.
+    # During batch sampling, the sampler ensures uniform distribution over the categories
+    # of the sampled sequences.
+    category_aware: bool = True
+
     seq_names: List[str] = field(init=False)
 
+    category_to_sequence_names: Dict[str, List[str]] = field(init=False)
+    categories: List[str] = field(init=False)
+
     def __post_init__(self) -> None:
         if self.batch_size <= 0:
             raise ValueError(
@@ -56,6 +66,10 @@ class SceneBatchSampler(Sampler[List[int]]):
 
         self.seq_names = list(self.dataset.sequence_names())
 
+        if self.category_aware:
+            self.category_to_sequence_names = self.dataset.category_to_sequence_names()
+            self.categories = list(self.category_to_sequence_names.keys())
+
     def __len__(self) -> int:
         return self.num_batches
 
@@ -67,7 +81,25 @@ class SceneBatchSampler(Sampler[List[int]]):
     def _sample_batch(self, batch_idx) -> List[int]:
         n_per_seq = np.random.choice(self.images_per_seq_options)
         n_seqs = -(-self.batch_size // n_per_seq)  # round up
-        chosen_seq = _capped_random_choice(self.seq_names, n_seqs, replace=False)
+
+        if self.category_aware:
+            # first sample categories at random, these can be repeated in the batch
+            chosen_cat = _capped_random_choice(self.categories, n_seqs, replace=True)
+            # then randomly sample a set of unique sequences within each category
+            chosen_seq = []
+            for cat, n_per_category in Counter(chosen_cat).items():
+                category_chosen_seq = _capped_random_choice(
+                    self.category_to_sequence_names[cat],
+                    n_per_category,
+                    replace=False,
+                )
+                chosen_seq.extend([str(s) for s in category_chosen_seq])
+        else:
+            chosen_seq = _capped_random_choice(
+                self.seq_names,
+                n_seqs,
+                replace=False,
+            )
 
         if self.sample_consecutive_frames:
             frame_idx = []

tests/implicitron/data/data_source.yaml

@@ -49,6 +49,7 @@ dataset_map_provider_JsonIndexDatasetMapProviderV2_args:
   test_on_train: false
   only_test_set: false
   load_eval_batches: true
+  num_load_workers: 4
   n_known_frames_for_test: 0
   dataset_class_type: JsonIndexDataset
   path_manager_factory_class_type: PathManagerFactory

tests/implicitron/test_batch_sampler.py

@@ -11,17 +11,20 @@ from dataclasses import dataclass
 from itertools import product
 
 import numpy as np
+
+import torch
 from pytorch3d.implicitron.dataset.data_loader_map_provider import (
     DoublePoolBatchSampler,
 )
 
-from pytorch3d.implicitron.dataset.dataset_base import DatasetBase
+from pytorch3d.implicitron.dataset.dataset_base import DatasetBase, FrameData
 from pytorch3d.implicitron.dataset.scene_batch_sampler import SceneBatchSampler
 
 
 @dataclass
 class MockFrameAnnotation:
     frame_number: int
+    sequence_name: str = "sequence"
     frame_timestamp: float = 0.0
 
 
@@ -41,6 +44,9 @@ class MockDataset(DatasetBase):
         self.frame_annots = [
             {"frame_annotation": MockFrameAnnotation(no)} for no in frame_nos * num_seq
         ]
+        for seq_name, idx in self._seq_to_idx.items():
+            for i in idx:
+                self.frame_annots[i]["frame_annotation"].sequence_name = seq_name
 
     def get_frame_numbers_and_timestamps(self, idxs):
         out = []
@@ -51,6 +57,16 @@ class MockDataset(DatasetBase):
             )
         return out
 
+    def __getitem__(self, index: int):
+        fa = self.frame_annots[index]["frame_annotation"]
+        fd = FrameData(
+            sequence_name=fa.sequence_name,
+            sequence_category="default_category",
+            frame_number=torch.LongTensor([fa.frame_number]),
+            frame_timestamp=torch.LongTensor([fa.frame_timestamp]),
+        )
+        return fd
+
 
 class TestSceneBatchSampler(unittest.TestCase):
     def setUp(self):

tests/implicitron/test_json_index_dataset_provider_v2.py

@@ -41,22 +41,73 @@ class TestJsonIndexDatasetProviderV2(unittest.TestCase):
         categories = ["A", "B"]
         subset_name = "test"
         eval_batch_size = 5
+        n_frames = 8 * 3
+        n_sequences = 5
+        n_eval_batches = 10
         with tempfile.TemporaryDirectory() as tmpd:
             _make_random_json_dataset_map_provider_v2_data(
                 tmpd,
                 categories,
                 eval_batch_size=eval_batch_size,
+                n_frames=n_frames,
+                n_sequences=n_sequences,
+                n_eval_batches=n_eval_batches,
             )
             for n_known_frames_for_test in [0, 2]:
-                for category in categories:
-                    dataset_provider = JsonIndexDatasetMapProviderV2(
+                dataset_providers = {
+                    category: JsonIndexDatasetMapProviderV2(
                         category=category,
                         subset_name="test",
                         dataset_root=tmpd,
                         n_known_frames_for_test=n_known_frames_for_test,
                     )
+                    for category in [*categories, ",".join(sorted(categories))]
+                }
+                for category, dataset_provider in dataset_providers.items():
                     dataset_map = dataset_provider.get_dataset_map()
                     for set_ in ["train", "val", "test"]:
+                        dataset = getattr(dataset_map, set_)
+
+                        cat2seq = dataset.category_to_sequence_names()
+                        self.assertEqual(",".join(sorted(cat2seq.keys())), category)
+
+                        if not (n_known_frames_for_test != 0 and set_ == "test"):
+                            # check the lengths only in case we do not have the
+                            # n_known_frames_for_test set
+                            expected_dataset_len = n_frames * n_sequences // 3
+                            if "," in category:
+                                # multicategory json index dataset, sum the lengths of
+                                # category-specific ones
+                                expected_dataset_len = sum(
+                                    len(
+                                        getattr(
+                                            dataset_providers[c].get_dataset_map(), set_
+                                        )
+                                    )
+                                    for c in categories
+                                )
+                                self.assertEqual(
+                                    sum(len(s) for s in cat2seq.values()),
+                                    n_sequences * len(categories),
+                                )
+                                self.assertEqual(len(cat2seq), len(categories))
+                            else:
+                                self.assertEqual(
+                                    len(cat2seq[category]),
+                                    n_sequences,
+                                )
+                                self.assertEqual(len(cat2seq), 1)
+                            self.assertEqual(len(dataset), expected_dataset_len)
+
+                        if set_ == "test":
+                            # check the number of eval batches
+                            expected_n_eval_batches = n_eval_batches
+                            if "," in category:
+                                expected_n_eval_batches *= len(categories)
+                            self.assertTrue(
+                                len(dataset.get_eval_batches())
+                                == expected_n_eval_batches
+                            )
                         if set_ in ["train", "val"]:
                             dataloader = torch.utils.data.DataLoader(
                                 getattr(dataset_map, set_),
@@ -80,6 +131,7 @@ class TestJsonIndexDatasetProviderV2(unittest.TestCase):
                         dataset_provider.get_category_to_subset_name_list()
                     )
                     category_to_subset_list_ = {c: [subset_name] for c in categories}
+
                     self.assertTrue(category_to_subset_list == category_to_subset_list_)
 
 
@@ -88,6 +140,7 @@ def _make_random_json_dataset_map_provider_v2_data(
     categories: List[str],
     n_frames: int = 8,
     n_sequences: int = 5,
+    n_eval_batches: int = 10,
     H: int = 50,
     W: int = 30,
     subset_name: str = "test",
@@ -100,7 +153,7 @@ def _make_random_json_dataset_map_provider_v2_data(
         sequence_annotations = []
         frame_index = []
         for seq_i in range(n_sequences):
-            seq_name = str(seq_i)
+            seq_name = category + str(seq_i)
             for i in range(n_frames):
                 # generate and store image
                 imdir = os.path.join(root, category, seq_name, "images")
@@ -165,7 +218,8 @@ def _make_random_json_dataset_map_provider_v2_data(
             json.dump(set_list, f)
 
         eval_batches = [
-            random.sample(test_frame_index, eval_batch_size) for _ in range(10)
+            random.sample(test_frame_index, eval_batch_size)
+            for _ in range(n_eval_batches)
         ]
 
         eval_b_dir = os.path.join(root, category, "eval_batches")