def random_select(img_scales):
assert mmcv.is_list_of(img_scales, tuple)
scale_idx = np.random.randint(len(img_scales))
img_scale = img_scales[scale_idx]
return img_scale, scale_idx
def __init__(self,
ann_file,
img_prefix,
img_scale,
img_norm_cfg,
size_divisor=None,
proposal_file=None,
num_max_proposals=1000,
flip_ratio=0,
with_mask=True,
with_crowd=False,
with_label=True,
with_track=False,
extra_aug=None,
aug_ref_bbox_param=None,
resize_keep_ratio=True,
test_mode=False):
# prefix of images path
self.img_prefix = img_prefix
self.ann_file = ann_file
# (long_edge, short_edge) or [(long1, short1), (long2, short2), ...]
self.img_norm_cfg = dict(mean=[0, 0, 0], std=[1, 1, 1], to_rgb=True)
self.img_scales = img_scale if isinstance(img_scale,
list) else [img_scale]
assert mmcv.is_list_of(self.img_scales, tuple)
# load annotations
self.vid_infos = self.load_annotations(ann_file)
img_ids = []
for v, vid_info in enumerate(self.vid_infos):
for f, ann in enumerate(vid_info['ann_info']):
img_ids.append([v, f])
self.img_ids = img_ids
# filter images with no annotation during training
if not test_mode:
valid_inds = [
i for i, (v, f) in enumerate(self.img_ids)
if len(self.get_ann_info(v, f)['bboxes'])
]
self.img_ids = [self.img_ids[i] for i in valid_inds]
# normalization configs
self.img_norm_cfg = img_norm_cfg
# max proposals per image
self.num_max_proposals = num_max_proposals
# flip ratio
self.flip_ratio = flip_ratio
assert flip_ratio >= 0 and flip_ratio <= 1
# padding border to ensure the image size can be divided by
# size_divisor (used for FPN)
self.size_divisor = size_divisor
# with mask or not (reserved field, takes no effect)
self.with_mask = with_mask
# some datasets provide bbox annotations as ignore/crowd/difficult,
# if `with_crowd` is True, then these info is returned.
self.with_crowd = with_crowd
# with label is False for RPN
self.with_label = with_label
self.with_track = with_track
# params for augmenting bbox in the reference frame
self.aug_ref_bbox_param = aug_ref_bbox_param
# in test mode or not
self.test_mode = test_mode
# set group flag for the sampler
if not self.test_mode:
self._set_group_flag()
# transforms
self.img_transform = ImageTransform(size_divisor=self.size_divisor,
**self.img_norm_cfg)
self.bbox_transform = BboxTransform()
self.mask_transform = MaskTransform()
self.numpy2tensor = Numpy2Tensor()
# if use extra augmentation
if extra_aug is not None:
self.extra_aug = ExtraAugmentation(**extra_aug)
else:
self.extra_aug = None
# image rescale if keep ratio
self.resize_keep_ratio = resize_keep_ratio
def __init__(self,
ann_file,
img_prefix,
img_scale,
img_norm_cfg,
multiscale_mode='value',
size_divisor=None,
proposal_file=None,
num_max_proposals=1000,
flip_ratio=0,
with_mask=True,
with_crowd=True,
with_label=True,
with_semantic_seg=False,
seg_prefix=None,
seg_scale_factor=1,
extra_aug=None,
rotate_aug=None,
rotate_test_aug=None,
resize_keep_ratio=True,
test_mode=False):
# prefix of images path
self.img_prefix = img_prefix
# load annotations (and proposals)
self.img_infos = self.load_annotations(ann_file)
if proposal_file is not None:
self.proposals = self.load_proposals(proposal_file)
else:
self.proposals = None
# filter images with no annotation during training
if not test_mode:
valid_inds = self._filter_imgs()
self.img_infos = [self.img_infos[i] for i in valid_inds]
if self.proposals is not None:
self.proposals = [self.proposals[i] for i in valid_inds]
# (long_edge, short_edge) or [(long1, short1), (long2, short2), ...]
self.img_scales = img_scale if isinstance(img_scale,
list) else [img_scale]
assert mmcv.is_list_of(self.img_scales, tuple)
# normalization configs
self.img_norm_cfg = img_norm_cfg
# multi-scale mode (only applicable for multi-scale training)
self.multiscale_mode = multiscale_mode
assert multiscale_mode in ['value', 'range']
# max proposals per image
self.num_max_proposals = num_max_proposals
# flip ratio
self.flip_ratio = flip_ratio
assert flip_ratio >= 0 and flip_ratio <= 1
# padding border to ensure the image size can be divided by
# size_divisor (used for FPN)
self.size_divisor = size_divisor
# with mask or not (reserved field, takes no effect)
self.with_mask = with_mask
# some datasets provide bbox annotations as ignore/crowd/difficult,
# if `with_crowd` is True, then these info is returned.
self.with_crowd = with_crowd
# with label is False for RPN
self.with_label = with_label
# with semantic segmentation (stuff) annotation or not
self.with_seg = with_semantic_seg
# prefix of semantic segmentation map path
self.seg_prefix = seg_prefix
# rescale factor for segmentation maps
self.seg_scale_factor = seg_scale_factor
# in test mode or not
self.test_mode = test_mode
# set group flag for the sampler
if not self.test_mode:
self._set_group_flag()
# transforms
self.img_transform = ImageTransform(size_divisor=self.size_divisor,
**self.img_norm_cfg)
self.bbox_transform = BboxTransform()
self.mask_transform = MaskTransform()
self.seg_transform = SegMapTransform(self.size_divisor)
self.numpy2tensor = Numpy2Tensor()
# if use extra augmentation
if extra_aug is not None:
self.extra_aug = ExtraAugmentation(**extra_aug)
else:
self.extra_aug = None
# if use rotation augmentation
if rotate_aug is not None:
self.rotate_aug = RotateAugmentation(self.CLASSES, **rotate_aug)
else:
self.rotate_aug = None
if rotate_test_aug is not None:
# dot not support argument settings currently
self.rotate_test_aug = RotateTestAugmentation()
else:
self.rotate_test_aug = None
# image rescale if keep ratio
self.resize_keep_ratio = resize_keep_ratio
def __init__(self,
ann_file,
img_prefix,
img_scale,
img_norm_cfg,
size_divisor=None,
proposal_file=None,
num_max_proposals=1000,
flip_ratio=0,
with_mask=True,
with_crowd=True,
with_label=True,
test_mode=False,
debug=False):
# path of the data file
self.coco = COCO(ann_file)
# filter images with no annotation during training
if not test_mode:
self.img_ids, self.img_infos = self._filter_imgs()
else:
self.img_ids = self.coco.getImgIds()
self.img_infos = [
self.coco.loadImgs(idx)[0] for idx in self.img_ids
]
assert len(self.img_ids) == len(self.img_infos)
# get the mapping from original category ids to labels
self.cat_ids = self.coco.getCatIds()
self.cat2label = {
cat_id: i + 1
for i, cat_id in enumerate(self.cat_ids)
}
# prefix of images path
self.img_prefix = img_prefix
# (long_edge, short_edge) or [(long1, short1), (long2, short2), ...]
self.img_scales = img_scale if isinstance(img_scale,
list) else [img_scale]
assert mmcv.is_list_of(self.img_scales, tuple)
# color channel order and normalize configs
self.img_norm_cfg = img_norm_cfg
# proposals
# TODO: revise _filter_imgs to be more flexible
if proposal_file is not None:
self.proposals = mmcv.load(proposal_file)
ori_ids = self.coco.getImgIds()
sorted_idx = [ori_ids.index(id) for id in self.img_ids]
self.proposals = [self.proposals[idx] for idx in sorted_idx]
else:
self.proposals = None
self.num_max_proposals = num_max_proposals
# flip ratio
self.flip_ratio = flip_ratio
assert flip_ratio >= 0 and flip_ratio <= 1
# padding border to ensure the image size can be divided by
# size_divisor (used for FPN)
self.size_divisor = size_divisor
# with crowd or not, False when using RetinaNet
self.with_crowd = with_crowd
# with mask or not
self.with_mask = with_mask
# with label is False for RPN
self.with_label = with_label
# in test mode or not
self.test_mode = test_mode
# debug mode or not
self.debug = debug
# set group flag for the sampler
self._set_group_flag()
# transforms
self.img_transform = ImageTransform(size_divisor=self.size_divisor,
**self.img_norm_cfg)
self.bbox_transform = BboxTransform()
self.mask_transform = MaskTransform()
self.numpy2tensor = Numpy2Tensor()
def run(self, data_loaders, workflow, max_epochs, **kwargs):
"""Start running.
Args:
data_loaders (list[:obj:`DataLoader`]): Dataloaders for training
and validation.
workflow (list[tuple]): A list of (phase, epochs) to specify the
running order and epochs. E.g, [('train', 2), ('val', 1)] means
running 2 epochs for training and 1 epoch for validation,
iteratively.
max_epochs (int): Total training epochs.
"""
not_done = True
max_retry = 10
max_retry_counter = 0
try:
self.pretrain_mode = kwargs['supcon_pretraining']
except:
self.pretrain_mode = False
kwargs = {k: v for k, v in kwargs.items() if k != 'supcon_pretraining'}
# Freeze the encoder if needed
if self.freeze_encoder:
for param in self.model.module.encoder.parameters():
param.requires_grad = False
while not_done:
print(
'===================starting training...========================='
)
# Reset the epoch counters
self.mode = None
self._epoch = 0
self._iter = 0
self._inner_iter = 0
self._max_epochs = 0
self._max_iters = 0
try:
print("Starting training for ", self.work_dir)
assert isinstance(data_loaders, list)
assert mmcv.is_list_of(workflow, tuple)
assert len(data_loaders) == len(workflow)
if self.pretrain_mode:
es_checkpoint = self.work_dir + '/pretrain_checkpoint.pt'
else:
es_checkpoint = self.work_dir + '/checkpoint.pt'
self.es_checkpoint = es_checkpoint
if self.early_stopping:
self.early_stopping_obj = EarlyStopping(
patience=self.es_patience,
verbose=True,
path=es_checkpoint)
self._max_epochs = max_epochs
for i, flow in enumerate(workflow):
mode, epochs = flow
if mode == 'train':
self._max_iters = self._max_epochs * len(
data_loaders[i])
break
work_dir = self.work_dir if self.work_dir is not None else 'NONE'
self.logger.info('Start running, host: %s, work_dir: %s',
get_host_info(), work_dir)
self.logger.info('workflow: %s, max: %d epochs', workflow,
max_epochs)
self.call_hook('before_run')
print("work dir: ", self.work_dir)
train_accs = np.zeros((1, max_epochs)) * np.nan
val_accs = np.zeros((1, max_epochs)) * np.nan
columns = ['epoch', 'train_acc', 'val_acc']
df_all = pd.DataFrame(columns=columns)
while self.epoch < max_epochs:
for i, flow in enumerate(workflow):
mode, epochs = flow
kwargs['workflow_stage'] = mode
kwargs['cur_epoch'] = self.epoch
if isinstance(mode, str): # self.train()
if not hasattr(self, mode):
raise ValueError(
f'runner has no method named "{mode}" to run an '
'epoch')
epoch_runner = getattr(self, mode)
elif callable(mode): # custom train()
epoch_runner = mode
else:
raise TypeError(
'mode in workflow must be a str or '
f'callable function, not {type(mode)}')
for _ in range(epochs):
if mode == 'train' and self.epoch >= max_epochs:
return
true_labels, predicted_labels = epoch_runner(
data_loaders[i], **kwargs)
if not self.pretrain_mode:
acc = accuracy_score(true_labels,
predicted_labels)
if mode == 'train':
df_all.loc[len(df_all)] = [
self.epoch - 1, acc,
val_accs[0, self.epoch - 1]
]
elif mode == 'val':
val_accs[0, self.epoch - 1] = acc
df_all.loc[df_all['epoch'] == self.epoch -
1, 'val_acc'] = acc
else:
# What to do after epoch if we're in pretrain mode
pass
if not self.pretrain_mode:
mmcv.mkdir_or_exist(self.work_dir)
df_all.to_csv(self.work_dir + "/results_df.csv")
if self.early_stopping:
print('checking early stopping:',
self.early_stopping_obj.early_stop,
self.force_run_all_epochs)
if not self.force_run_all_epochs and self.early_stopping_obj.early_stop:
print('should STOP now')
break
# We have successfully finished this participant
not_done = False
except Exception as e:
not_done = True
logging.exception(
"Error message ================================================="
)
max_retry_counter += 1
# Reset the model parameters
print(
"======================================going to retrain again, resetting parameters..."
)
print("This is the error we got:", e)
try:
self.model.module.apply(weight_reset)
if os.path.isfile(self.es_checkpoint):
os.remove(self.es_checkpoint)
print('deleted pretrain')
print('successfully reset weights')
except Exception as e:
print("This is the error we got _ 2:", e)
# Need to delete the wandb and work_dir folder to avoid filling up the cluster with files
# try:
# shutil.rmtree(wandb.run.dir)
# except:
# print('failed to delete the wandb folder')
try:
shutil.rmtree(self.work_dir)
except:
print('failed to delete the self.work_dir folder')
if max_retry_counter >= max_retry:
not_done = False
raise TooManyRetriesException
# If we stopped early, evaluate the performance of the saved model on all datasets
if self.early_stopping:
try:
self.log_buffer.update(
{'early_stop_epoch': self.early_stopping_epoch}, 1)
print('stopped at epoch: ', self.early_stopping_epoch)
except:
print(
"didn't meet early stopping criterion so we ran for all epochs"
)
print("*****************************now doing eval: ")
print("workflow", workflow)
print("data_loaders", data_loaders)
if not self.pretrain_mode:
self.early_stop_eval(es_checkpoint, workflow, data_loaders,
**kwargs)
else:
self.early_stop_eval_pretrain(es_checkpoint, workflow,
data_loaders, **kwargs)
time.sleep(10) # wait for some hooks like loggers to finish
self.call_hook('after_run')
return self.model, self.early_stopping_epoch
def evaluate_obs(self,
results,
metric='mIoU',
logger=None,
efficient_test=False,
**kwargs):
"""Evaluate the dataset.
Args:
results (list): Testing results of the dataset.
metric (str | list[str]): Metrics to be evaluated. 'mIoU' and
'mDice' are supported.
logger (logging.Logger | None | str): Logger used for printing
related information during evaluation. Default: None.
Returns:
dict[str, float]: Default metrics.
"""
if isinstance(metric, str):
metric = [metric]
allowed_metrics = ['mIoU', 'mDice']
if not set(metric).issubset(set(allowed_metrics)):
raise KeyError('metric {} is not supported'.format(metric))
eval_results = {}
gt_seg_maps = self.get_gt_seg_maps(efficient_test)
if self.CLASSES is None:
num_classes = len(
reduce(np.union1d, [np.unique(_) for _ in gt_seg_maps]))
else:
num_classes = len(self.CLASSES)
ret_metrics = eval_metrics(results,
gt_seg_maps,
num_classes,
self.ignore_index,
metric,
label_map=self.label_map,
reduce_zero_label=self.reduce_zero_label)
class_table_data = [['Class'] + [m[1:] for m in metric] + ['Acc']]
if self.CLASSES is None:
class_names = tuple(range(num_classes))
else:
class_names = self.CLASSES
ret_metrics_round = [
np.round(ret_metric * 100, 2) for ret_metric in ret_metrics
]
for i in range(num_classes):
class_table_data.append([class_names[i]] +
[m[i] for m in ret_metrics_round[2:]] +
[ret_metrics_round[1][i]])
summary_table_data = [['Scope'] +
['m' + head
for head in class_table_data[0][1:]] + ['aAcc']]
ret_metrics_mean = [
np.round(np.nanmean(ret_metric) * 100, 2)
for ret_metric in ret_metrics
]
summary_table_data.append(['global'] + ret_metrics_mean[2:] +
[ret_metrics_mean[1]] +
[ret_metrics_mean[0]])
print_log('per class results:', logger)
table = AsciiTable(class_table_data)
print_log('\n' + table.table, logger=logger)
print_log('Summary:', logger)
table = AsciiTable(summary_table_data)
print_log('\n' + table.table, logger=logger)
for i in range(1, len(summary_table_data[0])):
eval_results[summary_table_data[0]
[i]] = summary_table_data[1][i] / 100.0
if mmcv.is_list_of(results, str):
for file_name in results:
os.remove(file_name)
return eval_results
def __init__(self,
ann_file,
img_prefix,
img_scale,
img_norm_cfg,
size_divisor=None,
proposal_file=None,
num_max_proposals=1000,
flip_ratio=0,
with_mask=True,
with_crowd=True,
with_label=True,
test_mode=False):
# load annotations (and proposals)
self.img_infos = self.load_annotations(ann_file)
if proposal_file is not None:
self.proposals = self.load_proposals(proposal_file)
else:
self.proposals = None
# filter images with no annotation during training
if not test_mode:
valid_inds = self._filter_imgs()
self.img_infos = [self.img_infos[i] for i in valid_inds]
if self.proposals is not None:
self.proposals = [self.proposals[i] for i in valid_inds]
# prefix of images path
self.img_prefix = img_prefix
# (long_edge, short_edge) or [(long1, short1), (long2, short2), ...]
self.img_scales = img_scale if isinstance(img_scale,
list) else [img_scale]
assert mmcv.is_list_of(self.img_scales, tuple)
# normalization configs
self.img_norm_cfg = img_norm_cfg
# max proposals per image
self.num_max_proposals = num_max_proposals
# flip ratio
self.flip_ratio = flip_ratio
assert flip_ratio >= 0 and flip_ratio <= 1
# padding border to ensure the image size can be divided by
# size_divisor (used for FPN)
self.size_divisor = size_divisor
# with mask or not (reserved field, takes no effect)
self.with_mask = with_mask
# some datasets provide bbox annotations as ignore/crowd/difficult,
# if `with_crowd` is True, then these info is returned.
self.with_crowd = with_crowd
# with label is False for RPN
self.with_label = with_label
# in test mode or not
self.test_mode = test_mode
# set group flag for the sampler
if not self.test_mode:
self._set_group_flag()
# transforms
self.img_transform = ImageTransform(
size_divisor=self.size_divisor, **self.img_norm_cfg)
self.bbox_transform = BboxTransform()
self.mask_transform = MaskTransform()
self.numpy2tensor = Numpy2Tensor()
self.light=A.Compose([A.RGBShift(),A.InvertImg(),
A.Blur(),
A.GaussNoise(),
A.Flip(),
A.RandomRotate90()], bbox_params={'format':'pascal_voc','min_visibility': 0.4, 'label_fields': ['category_id']}, p=1)
def evaluate(self,
results,
metric='mIoU',
logger=None,
efficient_test=False,
**kwargs):
"""Evaluate the dataset.
Args:
results (list): Testing results of the dataset.
metric (str | list[str]): Metrics to be evaluated. 'mIoU',
'mDice' and 'mFscore' are supported.
logger (logging.Logger | None | str): Logger used for printing
related information during evaluation. Default: None.
Returns:
dict[str, float]: Default metrics.
"""
if isinstance(metric, str):
metric = [metric]
allowed_metrics = ['mIoU', 'mDice', 'mFscore']
if not set(metric).issubset(set(allowed_metrics)):
raise KeyError('metric {} is not supported'.format(metric))
eval_results = {}
gt_seg_maps = self.get_gt_seg_maps(efficient_test)
if self.CLASSES is None:
num_classes = len(
reduce(np.union1d, [np.unique(_) for _ in gt_seg_maps]))
else:
num_classes = len(self.CLASSES)
ret_metrics = eval_metrics(results,
gt_seg_maps,
num_classes,
self.ignore_index,
metric,
label_map=self.label_map,
reduce_zero_label=self.reduce_zero_label)
if self.CLASSES is None:
class_names = tuple(range(num_classes))
else:
class_names = self.CLASSES
# summary table
ret_metrics_summary = OrderedDict({
ret_metric: np.round(np.nanmean(ret_metric_value) * 100, 2)
for ret_metric, ret_metric_value in ret_metrics.items()
})
# each class table
ret_metrics.pop('aAcc', None)
ret_metrics_class = OrderedDict({
ret_metric: np.round(ret_metric_value * 100, 2)
for ret_metric, ret_metric_value in ret_metrics.items()
})
ret_metrics_class.update({'Class': class_names})
ret_metrics_class.move_to_end('Class', last=False)
# for logger
class_table_data = PrettyTable()
for key, val in ret_metrics_class.items():
class_table_data.add_column(key, val)
summary_table_data = PrettyTable()
for key, val in ret_metrics_summary.items():
if key == 'aAcc':
summary_table_data.add_column(key, [val])
else:
summary_table_data.add_column('m' + key, [val])
print_log('per class results:', logger)
print_log('\n' + class_table_data.get_string(), logger=logger)
print_log('Summary:', logger)
print_log('\n' + summary_table_data.get_string(), logger=logger)
# each metric dict
for key, value in ret_metrics_summary.items():
if key == 'aAcc':
eval_results[key] = value / 100.0
else:
eval_results['m' + key] = value / 100.0
ret_metrics_class.pop('Class', None)
for key, value in ret_metrics_class.items():
eval_results.update({
key + '.' + str(name): value[idx] / 100.0
for idx, name in enumerate(class_names)
})
if mmcv.is_list_of(results, str):
for file_name in results:
os.remove(file_name)
return eval_results
def __init__(self, transforms, img_scale, flip=False):
self.transforms = Compose(transforms)
self.img_scale = img_scale if isinstance(img_scale,
list) else [img_scale]
assert mmcv.is_list_of(self.img_scale, tuple)
self.flip = flip
def __init__(self,
ann_file,
img_prefix,
img_scale,
img_norm_cfg,
size_divisor=None,
proposal_file=None,
num_max_proposals=1000,
flip_ratio=0,
with_mask=True,
with_crowd=True,
with_label=True,
extra_aug=None,
resize_keep_ratio=True,
test_mode=False):
# prefix of images path
self.img_prefix = img_prefix
# load annotations (and proposals)
self.img_infos = self.load_annotations(ann_file)
if proposal_file is not None:
self.proposals = self.load_proposals(proposal_file)
else:
self.proposals = None
# filter images with no annotation during training
if not test_mode:
valid_inds = self._filter_imgs()
self.img_infos = [self.img_infos[i] for i in valid_inds]
if self.proposals is not None:
self.proposals = [self.proposals[i] for i in valid_inds]
# (long_edge, short_edge) or [(long1, short1), (long2, short2), ...]
self.img_scales = img_scale if isinstance(img_scale,
list) else [img_scale]
assert mmcv.is_list_of(self.img_scales, tuple)
# normalization configs
self.img_norm_cfg = img_norm_cfg
# max proposals per image
self.num_max_proposals = num_max_proposals
# flip ratio
self.flip_ratio = flip_ratio
assert flip_ratio >= 0 and flip_ratio <= 1
# padding border to ensure the image size can be divided by
# size_divisor (used for FPN)
self.size_divisor = size_divisor
# with mask or not (reserved field, takes no effect)
self.with_mask = with_mask
# some datasets provide bbox annotations as ignore/crowd/difficult,
# if `with_crowd` is True, then these info is returned.
self.with_crowd = with_crowd
# with label is False for RPN
self.with_label = with_label
# in test mode or not
self.test_mode = test_mode
# set group flag for the sampler
if not self.test_mode:
self._set_group_flag()
# transforms
self.img_transform = ImageTransform(size_divisor=self.size_divisor,
**self.img_norm_cfg)
self.bbox_transform = BboxTransform()
self.mask_transform = MaskTransform()
self.numpy2tensor = Numpy2Tensor()
# if use extra augmentation
if extra_aug is not None:
self.extra_aug = ExtraAugmentation(**extra_aug)
else:
self.extra_aug = None
# image rescale if keep ratio
self.resize_keep_ratio = resize_keep_ratio
# 注意这里的对应label是从1开始(1-20)，所以如果做逆对应就需要-1才能得到对应的真实描述
self.cat2label = {cat: i + 1 for i, cat in enumerate(self.CLASSES)}
if 'VOC2007' in self.img_prefix:
self.year = 2007
elif 'VOC2012' in self.img_prefix:
self.year = 2012
elif 'VOC2007' in self.img_prefix and 'VOC2012' in self.img_prefix:
self.year = 1207
else:
raise ValueError('Cannot infer dataset year from img_prefix')
def __init__(self,
ann_file,
img_prefix,
img_scale,
img_norm_cfg,
multiscale_mode='value',
size_divisor=None,
proposal_file=None,
num_max_proposals=1000,
flip_ratio=0,
with_mask=True,
with_crowd=True,
with_label=True,
with_semantic_seg=False,
seg_prefix=None,
seg_scale_factor=1,
extra_aug=None,
resize_keep_ratio=True,
test_mode=False,
remove_small_box=False,
small_box_size=8,
strides=None,
regress_ranges=None,
upper_factor=None,
upper_more_factor=None,
with_width=False):
# prefix of images path
self.img_prefix = img_prefix
# load annotations (and proposals)
self.img_infos = self.load_annotations(ann_file)
if proposal_file is not None:
self.proposals = self.load_proposals(proposal_file)
else:
self.proposals = None
# filter images with no annotation during training
if not test_mode:
valid_inds = self._filter_imgs()
self.img_infos = [self.img_infos[i] for i in valid_inds]
if self.proposals is not None:
self.proposals = [self.proposals[i] for i in valid_inds]
# (long_edge, short_edge) or [(long1, short1), (long2, short2), ...]
self.img_scales = img_scale if isinstance(img_scale,
list) else [img_scale]
assert mmcv.is_list_of(self.img_scales, tuple)
# normalization configs
self.img_norm_cfg = img_norm_cfg
# multi-scale mode (only applicable for multi-scale training)
self.multiscale_mode = multiscale_mode
assert multiscale_mode in ['value', 'range']
# max proposals per image
self.num_max_proposals = num_max_proposals
# flip ratio
self.flip_ratio = flip_ratio
assert flip_ratio >= 0 and flip_ratio <= 1
# padding border to ensure the image size can be divided by
# size_divisor (used for FPN)
self.size_divisor = size_divisor
# with mask or not (reserved field, takes no effect)
self.with_mask = with_mask
# some datasets provide bbox annotations as ignore/crowd/difficult,
# if `with_crowd` is True, then these info is returned.
self.with_crowd = with_crowd
# with label is False for RPN
self.with_label = with_label
# with semantic segmentation (stuff) annotation or not
self.with_seg = with_semantic_seg
# prefix of semantic segmentation map path
self.seg_prefix = seg_prefix
# rescale factor for segmentation maps
self.seg_scale_factor = seg_scale_factor
# in test mode or not
self.test_mode = test_mode
# remove small size box in gt
self.remove_small_box = remove_small_box
# the smallest box edge
self.small_box_size = small_box_size
# strides of FPN style outputs, used for generating groundtruth feature maps
self.strides = strides
# regress range of FPN style outputs, used for generating groundtruth feature maps
self.regress_ranges = regress_ranges
# upper factor for Irtiza's model which use three branches to predict upper box, full box, lower box
self.upper_factor = upper_factor
# split the upper box into more boxes
self.upper_more_factor = upper_more_factor
# set group flag for the sampler
if not self.test_mode:
self._set_group_flag()
# transforms
self.img_transform = ImageTransform(size_divisor=self.size_divisor,
**self.img_norm_cfg)
self.bbox_transform = BboxTransform()
self.mask_transform = MaskTransform()
self.seg_transform = SegMapTransform(self.size_divisor)
self.numpy2tensor = Numpy2Tensor()
# if use extra augmentation
if extra_aug is not None:
self.extra_aug = ExtraAugmentation(**extra_aug)
else:
self.extra_aug = None
# image rescale if keep ratio
self.resize_keep_ratio = resize_keep_ratio
#predict width
self.with_width = with_width
def run(self, data_loaders, workflow, max_epochs=None, **kwargs):
"""Start running.
Args:
data_loaders (list[:obj:`DataLoader`]): Dataloaders for training
and validation.
workflow (list[tuple]): A list of (phase, epochs) to specify the
running order and epochs. E.g, [('train', 2), ('val', 1)] means
running 2 epochs for training and 1 epoch for validation,
iteratively.
"""
assert isinstance(data_loaders, list)
assert mmcv.is_list_of(workflow, tuple)
assert len(data_loaders) == len(workflow)
if max_epochs is not None:
warnings.warn(
'setting max_epochs in run is deprecated, '
'please set max_epochs in runner_config', DeprecationWarning)
self._max_epochs = max_epochs
assert self._max_epochs is not None, (
'max_epochs must be specified during instantiation')
for i, flow in enumerate(workflow):
mode, epochs = flow
if mode == 'train':
if isinstance(data_loaders[i], zip):
self._max_iters = self._max_epochs * kwargs[
'epoch_max_iters']
else:
self._max_iters = self._max_epochs * len(data_loaders[i])
break
work_dir = self.work_dir if self.work_dir is not None else 'NONE'
self.logger.info('Start running, host: %s, work_dir: %s',
get_host_info(), work_dir)
self.logger.info('workflow: %s, max: %d epochs', workflow,
self._max_epochs)
self.call_hook('before_run')
while self.epoch < self._max_epochs:
for i, flow in enumerate(workflow):
mode, epochs = flow
if isinstance(mode, str): # self.train()
if not hasattr(self, mode):
raise ValueError(
f'runner has no method named "{mode}" to run an '
'epoch')
epoch_runner = getattr(self, mode)
else:
raise TypeError(
'mode in workflow must be a str, but got {}'.format(
type(mode)))
for _ in range(epochs):
if mode == 'train' and self.epoch >= self._max_epochs:
break
epoch_runner(data_loaders[i], **kwargs)
time.sleep(1) # wait for some hooks like loggers to finish
self.call_hook('after_run')
