148 lines
5.5 KiB
Python
148 lines
5.5 KiB
Python
# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
|
# SPDX-License-Identifier: Apache-2.0
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
|
|
from typing import Dict, Union
|
|
|
|
import numpy as np
|
|
import torch
|
|
import webdataset
|
|
|
|
from chronoedit._ext.imaginaire.lazy_config import instantiate
|
|
|
|
|
|
class IterativeJointDataLoader(webdataset.WebLoader):
|
|
r"""
|
|
A joint dataloader that supports loading both images and videos.
|
|
"""
|
|
|
|
def __init__(
|
|
self, dataloaders: Dict[str, Dict[str, Union[torch.utils.data.DataLoader, webdataset.WebLoader, int]]]
|
|
):
|
|
"""
|
|
Initialize the JointDataLoader with multiple datasets.
|
|
|
|
Args:
|
|
dataloaders: key - dataset_name; value - {"dataloader": dataloader, "ratio": data_ratio}
|
|
|
|
Example:
|
|
joint_loader = IterativeJointDataLoader(
|
|
dataloaders{
|
|
"image_data": {
|
|
"dataloader": webdataset.WebLoader(...),
|
|
"ratio": 4,
|
|
},
|
|
"video_data": {
|
|
"dataloader": torch.utils.data.DataLoader(...),
|
|
"ratio": 1,
|
|
},
|
|
}
|
|
)
|
|
"""
|
|
self.dataloader_list, self.dataset_name_list, self.data_ratios = [], [], []
|
|
|
|
for dataset_name, dataloader_data in dataloaders.items():
|
|
assert set(dataloader_data.keys()) == {"dataloader", "ratio"}, f"Invalid config: {dataloader_data}"
|
|
self.dataset_name_list.append(dataset_name)
|
|
self.dataloader_list.append(instantiate(dataloader_data["dataloader"]))
|
|
self.data_ratios.append(dataloader_data["ratio"])
|
|
|
|
self.global_id = 0
|
|
self.ratio_sum = sum(self.data_ratios)
|
|
|
|
self.data_len = 0
|
|
self.dataloaders = [iter(dataloader) for dataloader in self.dataloader_list]
|
|
for data in self.dataloader_list:
|
|
self.data_len += len(data)
|
|
|
|
def __len__(self) -> int:
|
|
return self.data_len
|
|
|
|
def __iter__(self):
|
|
while True:
|
|
data_id = self.global_id % self.ratio_sum
|
|
index_id = self._get_dataloader_index(data_id)
|
|
curr_dataloader = self.dataloaders[index_id]
|
|
output = next(curr_dataloader)
|
|
output["dataset_name"] = self.dataset_name_list[index_id]
|
|
self.global_id += 1
|
|
del curr_dataloader
|
|
yield output
|
|
|
|
def _get_dataloader_index(self, data_id):
|
|
"""Maps global id to the corresponding dataloader index based on ratio."""
|
|
for i, r in enumerate(self.data_ratios):
|
|
if data_id < r:
|
|
return i
|
|
data_id -= r
|
|
raise ValueError("Invalid data_id")
|
|
|
|
|
|
class RandomJointDataLoader(webdataset.WebLoader):
|
|
r"""
|
|
A joint dataloader that supports randomly samples batches from multiple datasets.
|
|
"""
|
|
|
|
# def __init__(self, **kwargs):
|
|
def __init__(
|
|
self, dataloaders: Dict[str, Dict[str, Union[torch.utils.data.DataLoader, webdataset.WebLoader, int]]]
|
|
):
|
|
"""
|
|
Initialize the JointDataLoader with multiple datasets.
|
|
|
|
Args:
|
|
**kwargs: Arbitrary keyword arguments where each key is a string
|
|
representing the dataset name, and each value is either
|
|
a `webdataset.WebLoader` or `torch.utils.data.DataLoader`
|
|
instance.
|
|
|
|
Raises:
|
|
AssertionError: If any value in kwargs is not an instance of
|
|
`webdataset.WebLoader` or `torch.utils.data.DataLoader`.
|
|
AssertionError: If any key in kwargs is not a string.
|
|
|
|
Example:
|
|
joint_loader = JointDataLoader(
|
|
images=webdataset.WebLoader(...),
|
|
videos=torch.utils.data.DataLoader(...)
|
|
)
|
|
"""
|
|
self.dataloader_list, self.dataset_name_list, self.data_ratios = [], [], []
|
|
|
|
for dataset_name, dataloader_data in dataloaders.items():
|
|
assert set(dataloader_data.keys()) == {"dataloader", "ratio"}, f"Invalid config: {dataloader_data}"
|
|
self.dataset_name_list.append(dataset_name)
|
|
self.dataloader_list.append(instantiate(dataloader_data["dataloader"]))
|
|
self.data_ratios.append(dataloader_data["ratio"])
|
|
|
|
assert np.isclose(sum(self.data_ratios), 1.0), "Sum of sample probabilities should be equal to 1."
|
|
|
|
self.data_len = 0
|
|
self.dataloaders = [iter(dataloader) for dataloader in self.dataloader_list]
|
|
for data in self.dataloader_list:
|
|
self.data_len += len(data)
|
|
|
|
def __len__(self) -> int:
|
|
return self.data_len
|
|
|
|
def __iter__(self):
|
|
while True:
|
|
# Sample a random dataset
|
|
data_id = int(np.random.choice(len(self.dataloader_list), 1, p=self.data_ratios)[0])
|
|
curr_dataloader = self.dataloaders[data_id]
|
|
output = next(curr_dataloader)
|
|
output["dataset_name"] = self.dataset_name_list[data_id]
|
|
del curr_dataloader
|
|
yield output
|