import logging
import time
import weakref
from typing import Dict
import numpy as np
import torch
from cvpods.utils import comm
from cvpods.utils.dump.events import EventStorage, get_event_storage
from cvpods.utils.registry import Registry
from .hooks import HookBase
RUNNERS = Registry("runners")
logger = logging.getLogger(__name__)
@RUNNERS.register()
class RunnerBase:
"""
Base class for iterative runner with hooks.
The only assumption we made here is: the training runs in a loop.
A subclass can implement what the loop is.
We made no assumptions about the existence of dataloader, optimizer, model, etc.
Attributes:
iter(int): the current iteration.
#!/usr/bin/python3
# -*- coding: utf-8 -*-
# Copyright (C) 2019-2021 Megvii Inc. All rights reserved.
from torch.optim import lr_scheduler
from cvpods.utils.registry import Registry
from .lr_scheduler import PolyLR, WarmupCosineLR, WarmupMultiStepLR
SCHEDULER_BUILDER = Registry("LRScheduler builder")
@SCHEDULER_BUILDER.register()
class BaseSchedulerBuilder:
@staticmethod
def build(optimizer, cfg, **kwargs):
raise NotImplementedError
@SCHEDULER_BUILDER.register()
class WarmupMultiStepLRBuilder(BaseSchedulerBuilder):
@staticmethod
def build(optimizer, cfg, **kwargs):
scheduler = WarmupMultiStepLR(
optimizer,
cfg.SOLVER.LR_SCHEDULER.STEPS,
cfg.SOLVER.LR_SCHEDULER.GAMMA,
warmup_factor=cfg.SOLVER.LR_SCHEDULER.WARMUP_FACTOR,
warmup_iters=cfg.SOLVER.LR_SCHEDULER.WARMUP_ITERS,
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
# Copyright (c) BaseDetection, Inc. and its affiliates. All Rights Reserved
from typing import Any, Dict, List, Set
import torch
from torch import optim
from cvpods.utils.registry import Registry
OPTIMIZER_BUILDER = Registry("Optimizer builder")
NORM_MODULE_TYPES = (
torch.nn.BatchNorm1d,
torch.nn.BatchNorm2d,
torch.nn.BatchNorm3d,
torch.nn.SyncBatchNorm,
# NaiveSyncBatchNorm inherits from BatchNorm2d
torch.nn.GroupNorm,
torch.nn.InstanceNorm1d,
torch.nn.InstanceNorm2d,
torch.nn.InstanceNorm3d,
torch.nn.LayerNorm,
torch.nn.LocalResponseNorm,
)
@OPTIMIZER_BUILDER.register()
class OptimizerBuilder:
@staticmethod
import torch
from torch.optim import Optimizer
from torch.optim.lr_scheduler import _LRScheduler
from cvpods.utils.registry import Registry
OPTIMIZERS = Registry("optimizers")
LR_SCHEDULERS = Registry("lr_schedulers")
for attr in dir(torch.optim):
optim = getattr(torch.optim, attr)
try:
if issubclass(optim, Optimizer):
OPTIMIZERS.register(optim)
except:
continue
for attr in dir(torch.optim.lr_scheduler):
lrs = getattr(torch.optim.lr_scheduler, attr)
try:
if issubclass(lrs, _LRScheduler):
LR_SCHEDULERS.register(lrs)
except:
continue