mirror of https://github.com/facebookresearch/deit
60 lines
2.2 KiB
Python
60 lines
2.2 KiB
Python
|
# Copyright (c) 2015-present, Facebook, Inc.
|
||
|
|
# All rights reserved.
|
||
|
|
import torch
|
||
|
|
import torch.distributed as dist
|
||
|
|
import math
|
||
|
|
|
||
|
|
|
||
|
|
class RASampler(torch.utils.data.Sampler):
|
||
|
|
"""Sampler that restricts data loading to a subset of the dataset for distributed,
|
||
|
|
with repeated augmentation.
|
||
|
|
It ensures that different each augmented version of a sample will be visible to a
|
||
|
|
different process (GPU)
|
||
|
|
Heavily based on torch.utils.data.DistributedSampler
|
||
|
|
"""
|
||
|
|
|
||
|
|
def __init__(self, dataset, num_replicas=None, rank=None, shuffle=True):
|
||
|
|
if num_replicas is None:
|
||
|
|
if not dist.is_available():
|
||
|
|
raise RuntimeError("Requires distributed package to be available")
|
||
|
|
num_replicas = dist.get_world_size()
|
||
|
|
if rank is None:
|
||
|
|
if not dist.is_available():
|
||
|
|
raise RuntimeError("Requires distributed package to be available")
|
||
|
|
rank = dist.get_rank()
|
||
|
|
self.dataset = dataset
|
||
|
|
self.num_replicas = num_replicas
|
||
|
|
self.rank = rank
|
||
|
|
self.epoch = 0
|
||
|
|
self.num_samples = int(math.ceil(len(self.dataset) * 3.0 / self.num_replicas))
|
||
|
|
self.total_size = self.num_samples * self.num_replicas
|
||
|
|
# self.num_selected_samples = int(math.ceil(len(self.dataset) / self.num_replicas))
|
||
|
|
self.num_selected_samples = int(math.floor(len(self.dataset) // 256 * 256 / self.num_replicas))
|
||
|
|
self.shuffle = shuffle
|
||
|
|
|
||
|
|
def __iter__(self):
|
||
|
|
# deterministically shuffle based on epoch
|
||
|
|
g = torch.Generator()
|
||
|
|
g.manual_seed(self.epoch)
|
||
|
|
if self.shuffle:
|
||
|
|
indices = torch.randperm(len(self.dataset), generator=g).tolist()
|
||
|
|
else:
|
||
|
|
indices = list(range(len(self.dataset)))
|
||
|
|
|
||
|
|
# add extra samples to make it evenly divisible
|
||
|
|
indices = [ele for ele in indices for i in range(3)]
|
||
|
|
indices += indices[:(self.total_size - len(indices))]
|
||
|
|
assert len(indices) == self.total_size
|
||
|
|
|
||
|
|
# subsample
|
||
|
|
indices = indices[self.rank:self.total_size:self.num_replicas]
|
||
|
|
assert len(indices) == self.num_samples
|
||
|
|
|
||
|
|
return iter(indices[:self.num_selected_samples])
|
||
|
|
|
||
|
|
def __len__(self):
|
||
|
|
return self.num_selected_samples
|
||
|
|
|
||
|
|
def set_epoch(self, epoch):
|
||
|
|
self.epoch = epoch
|