You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
101 lines
4.1 KiB
101 lines
4.1 KiB
import torch
|
|
import torch.utils.model_zoo as model_zoo
|
|
import os
|
|
import logging
|
|
from collections import OrderedDict
|
|
|
|
|
|
def load_checkpoint(model, checkpoint_path, use_ema=False):
|
|
if checkpoint_path and os.path.isfile(checkpoint_path):
|
|
checkpoint = torch.load(checkpoint_path, map_location='cpu')
|
|
state_dict_key = ''
|
|
if isinstance(checkpoint, dict):
|
|
state_dict_key = 'state_dict'
|
|
if use_ema and 'state_dict_ema' in checkpoint:
|
|
state_dict_key = 'state_dict_ema'
|
|
if state_dict_key and state_dict_key in checkpoint:
|
|
new_state_dict = OrderedDict()
|
|
for k, v in checkpoint[state_dict_key].items():
|
|
# strip `module.` prefix
|
|
name = k[7:] if k.startswith('module') else k
|
|
new_state_dict[name] = v
|
|
model.load_state_dict(new_state_dict)
|
|
else:
|
|
model.load_state_dict(checkpoint)
|
|
logging.info("Loaded {} from checkpoint '{}'".format(state_dict_key or 'weights', checkpoint_path))
|
|
else:
|
|
logging.error("No checkpoint found at '{}'".format(checkpoint_path))
|
|
raise FileNotFoundError()
|
|
|
|
|
|
def resume_checkpoint(model, checkpoint_path):
|
|
other_state = {}
|
|
resume_epoch = None
|
|
if os.path.isfile(checkpoint_path):
|
|
checkpoint = torch.load(checkpoint_path, map_location='cpu')
|
|
if isinstance(checkpoint, dict) and 'state_dict' in checkpoint:
|
|
new_state_dict = OrderedDict()
|
|
for k, v in checkpoint['state_dict'].items():
|
|
name = k[7:] if k.startswith('module') else k
|
|
new_state_dict[name] = v
|
|
model.load_state_dict(new_state_dict)
|
|
if 'optimizer' in checkpoint:
|
|
other_state['optimizer'] = checkpoint['optimizer']
|
|
if 'amp' in checkpoint:
|
|
other_state['amp'] = checkpoint['amp']
|
|
if 'epoch' in checkpoint:
|
|
resume_epoch = checkpoint['epoch']
|
|
if 'version' in checkpoint and checkpoint['version'] > 1:
|
|
resume_epoch += 1 # start at the next epoch, old checkpoints incremented before save
|
|
logging.info("Loaded checkpoint '{}' (epoch {})".format(checkpoint_path, checkpoint['epoch']))
|
|
else:
|
|
model.load_state_dict(checkpoint)
|
|
logging.info("Loaded checkpoint '{}'".format(checkpoint_path))
|
|
return other_state, resume_epoch
|
|
else:
|
|
logging.error("No checkpoint found at '{}'".format(checkpoint_path))
|
|
raise FileNotFoundError()
|
|
|
|
|
|
def load_pretrained(model, cfg=None, num_classes=1000, in_chans=3, filter_fn=None):
|
|
if cfg is None:
|
|
cfg = getattr(model, 'default_cfg')
|
|
if cfg is None or 'url' not in cfg or not cfg['url']:
|
|
logging.warning("Pretrained model URL is invalid, using random initialization.")
|
|
return
|
|
|
|
state_dict = model_zoo.load_url(cfg['url'], progress=False, map_location='cpu')
|
|
|
|
if in_chans == 1:
|
|
conv1_name = cfg['first_conv']
|
|
logging.info('Converting first conv (%s) from 3 to 1 channel' % conv1_name)
|
|
conv1_weight = state_dict[conv1_name + '.weight']
|
|
state_dict[conv1_name + '.weight'] = conv1_weight.sum(dim=1, keepdim=True)
|
|
elif in_chans != 3:
|
|
assert False, "Invalid in_chans for pretrained weights"
|
|
|
|
strict = True
|
|
classifier_name = cfg['classifier']
|
|
if num_classes == 1000 and cfg['num_classes'] == 1001:
|
|
# special case for imagenet trained models with extra background class in pretrained weights
|
|
classifier_weight = state_dict[classifier_name + '.weight']
|
|
state_dict[classifier_name + '.weight'] = classifier_weight[1:]
|
|
classifier_bias = state_dict[classifier_name + '.bias']
|
|
state_dict[classifier_name + '.bias'] = classifier_bias[1:]
|
|
elif num_classes != cfg['num_classes']:
|
|
# completely discard fully connected for all other differences between pretrained and created model
|
|
del state_dict[classifier_name + '.weight']
|
|
del state_dict[classifier_name + '.bias']
|
|
strict = False
|
|
|
|
if filter_fn is not None:
|
|
state_dict = filter_fn(state_dict)
|
|
|
|
model.load_state_dict(state_dict, strict=strict)
|
|
|
|
|
|
|
|
|
|
|
|
|