HuQiong
/
PaddleRS
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge branch 'develop' into refactor_res

Browse Source
own
Bobholamovic 3 years ago
parent 9e6e6c717c 68443d06cd
commit 0533d41334
6 changed files with 92 additions and 290 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 16
      paddlers/deploy/predictor.py
  2. 12
      paddlers/tasks/change_detector.py
  3. 20
      paddlers/tasks/classifier.py
  4. 302
      paddlers/tasks/object_detector.py
  5. 20
      paddlers/tasks/segmenter.py
  6. 12
      test_tipc/infer.py

16
paddlers/deploy/predictor.py
Unescape View File

@ -146,7 +146,7 @@ class Predictor(object):
return predictor
def preprocess(self, images, transforms):
preprocessed_samples = self._model._preprocess(
preprocessed_samples = self._model.preprocess(
images, transforms, to_tensor=False)
if self._model.model_type == 'classifier':
preprocessed_samples = {'image': preprocessed_samples[0]}
@ -182,12 +182,12 @@ class Predictor(object):
transforms=None):
if self._model.model_type == 'classifier':
true_topk = min(self._model.num_classes, topk)
if self._model._postprocess is None:
self._model.build_postprocess_from_labels(true_topk)
# XXX: Convert ndarray to tensor as self._model._postprocess requires
if self._model.postprocess is None:
self._model.build_postprocess_from_labels(topk)
# XXX: Convert ndarray to tensor as self._model.postprocess requires
assert len(net_outputs) == 1
net_outputs = paddle.to_tensor(net_outputs[0])
outputs = self._model._postprocess(net_outputs)
outputs = self._model.postprocess(net_outputs)
class_ids = map(itemgetter('class_ids'), outputs)
scores = map(itemgetter('scores'), outputs)
label_names = map(itemgetter('label_names'), outputs)
@ -197,7 +197,7 @@ class Predictor(object):
'label_names_map': n,
} for l, s, n in zip(class_ids, scores, label_names)]
elif self._model.model_type in ('segmenter', 'change_detector'):
label_map, score_map = self._model._postprocess(
label_map, score_map = self._model.postprocess(
net_outputs,
batch_origin_shape=ori_shape,
transforms=transforms.transforms)
@ -210,9 +210,9 @@ class Predictor(object):
k: v
for k, v in zip(['bbox', 'bbox_num', 'mask'], net_outputs)
}
preds = self._model._postprocess(net_outputs)
preds = self._model.postprocess(net_outputs)
elif self._model.model_type == 'restorer':
res_maps = self._model._postprocess(
res_maps = self._model.postprocess(
net_outputs[0],
batch_tar_shape=tar_shape,
transforms=transforms.transforms)

12
paddlers/tasks/change_detector.py
Unescape View File

@ -119,10 +119,10 @@ class BaseChangeDetector(BaseModel):
if mode == 'test':
origin_shape = inputs[2]
if self.status == 'Infer':
label_map_list, score_map_list = self._postprocess(
label_map_list, score_map_list = self.postprocess(
net_out, origin_shape, transforms=inputs[3])
else:
logit_list = self._postprocess(
logit_list = self.postprocess(
logit, origin_shape, transforms=inputs[3])
label_map_list = []
score_map_list = []
@ -150,7 +150,7 @@ class BaseChangeDetector(BaseModel):
raise ValueError("Expected label.ndim == 4 but got {}".format(
label.ndim))
origin_shape = [label.shape[-2:]]
pred = self._postprocess(
pred = self.postprocess(
pred, origin_shape, transforms=inputs[3])[0] # NCHW
intersect_area, pred_area, label_area = ppseg.utils.metrics.calculate_area(
pred, label, self.num_classes)
@ -556,7 +556,7 @@ class BaseChangeDetector(BaseModel):
images = [img_file]
else:
images = img_file
batch_im1, batch_im2, batch_origin_shape = self._preprocess(
batch_im1, batch_im2, batch_origin_shape = self.preprocess(
images, transforms, self.model_type)
self.net.eval()
data = (batch_im1, batch_im2, batch_origin_shape, transforms.transforms)
@ -667,7 +667,7 @@ class BaseChangeDetector(BaseModel):
dst_data = None
print("GeoTiff saved in {}.".format(save_file))
def _preprocess(self, images, transforms, to_tensor=True):
def preprocess(self, images, transforms, to_tensor=True):
self._check_transforms(transforms, 'test')
batch_im1, batch_im2 = list(), list()
batch_ori_shape = list()
@ -739,7 +739,7 @@ class BaseChangeDetector(BaseModel):
batch_restore_list.append(restore_list)
return batch_restore_list
def _postprocess(self, batch_pred, batch_origin_shape, transforms):
def postprocess(self, batch_pred, batch_origin_shape, transforms):
batch_restore_list = BaseChangeDetector.get_transforms_shape_info(
batch_origin_shape, transforms)
if isinstance(batch_pred, (tuple, list)) and self.status == 'Infer':

20
paddlers/tasks/classifier.py
Unescape View File

@ -62,7 +62,7 @@ class BaseClassifier(BaseModel):
self.metrics = None
self.losses = losses
self.labels = None
self._postprocess = None
self.postprocess = None
if params.get('with_net', True):
params.pop('with_net', None)
self.net = self.build_net(**params)
@ -127,13 +127,12 @@ class BaseClassifier(BaseModel):
net_out = net(inputs[0])
if mode == 'test':
return self._postprocess(net_out)
return self.postprocess(net_out)
outputs = OrderedDict()
label = paddle.to_tensor(inputs[1], dtype="int64")
if mode == 'eval':
# print(self._postprocess(net_out)[0]) # for test
label = paddle.unsqueeze(label, axis=-1)
metric_dict = self.metrics(net_out, label)
outputs['top1'] = metric_dict["top1"]
@ -182,13 +181,13 @@ class BaseClassifier(BaseModel):
label_dict = dict()
for i, label in enumerate(self.labels):
label_dict[i] = label
self._postprocess = build_postprocess({
self.postprocess = build_postprocess({
"name": "Topk",
"topk": topk,
"class_id_map_file": None
})
# Add class_id_map from model.yml
self._postprocess.class_id_map = label_dict
self.postprocess.class_id_map = label_dict
def train(self,
num_epochs,
@ -253,8 +252,7 @@ class BaseClassifier(BaseModel):
if self.losses is None:
self.losses = self.default_loss()
self.metrics = self.default_metric()
self._postprocess = self.default_postprocess(train_dataset.label_list)
# print(self._postprocess.class_id_map)
self.postprocess = self.default_postprocess(train_dataset.label_list)
if optimizer is None:
num_steps_each_epoch = train_dataset.num_samples // train_batch_size
@ -460,12 +458,12 @@ class BaseClassifier(BaseModel):
images = [img_file]
else:
images = img_file
batch_im, batch_origin_shape = self._preprocess(images, transforms,
self.model_type)
batch_im, batch_origin_shape = self.preprocess(images, transforms,
self.model_type)
self.net.eval()
data = (batch_im, batch_origin_shape, transforms.transforms)
if self._postprocess is None:
if self.postprocess is None:
self.build_postprocess_from_labels()
outputs = self.run(self.net, data, 'test')
@ -486,7 +484,7 @@ class BaseClassifier(BaseModel):
}
return prediction
def _preprocess(self, images, transforms, to_tensor=True):
def preprocess(self, images, transforms, to_tensor=True):
self._check_transforms(transforms, 'test')
batch_im = list()
batch_ori_shape = list()

302
paddlers/tasks/object_detector.py
Unescape View File

@ -254,6 +254,34 @@ class BaseDetector(BaseModel):
Defaults to None.
"""
args = self._pre_train(locals())
return self._real_train(**args)
def _pre_train(self, in_args):
return in_args
def _real_train(self,
num_epochs,
train_dataset,
train_batch_size=64,
eval_dataset=None,
optimizer=None,
save_interval_epochs=1,
log_interval_steps=10,
save_dir='output',
pretrain_weights='IMAGENET',
learning_rate=.001,
warmup_steps=0,
warmup_start_lr=0.0,
lr_decay_epochs=(216, 243),
lr_decay_gamma=0.1,
metric=None,
use_ema=False,
early_stop=False,
early_stop_patience=5,
use_vdl=True,
resume_checkpoint=None):
if self.status == 'Infer':
logging.error(
"Exported inference model does not support training.",
@ -583,16 +611,16 @@ class BaseDetector(BaseModel):
else:
images = img_file
batch_samples = self._preprocess(images, transforms)
batch_samples = self.preprocess(images, transforms)
self.net.eval()
outputs = self.run(self.net, batch_samples, 'test')
prediction = self._postprocess(outputs)
prediction = self.postprocess(outputs)
if isinstance(img_file, (str, np.ndarray)):
prediction = prediction[0]
return prediction
def _preprocess(self, images, transforms, to_tensor=True):
def preprocess(self, images, transforms, to_tensor=True):
self._check_transforms(transforms, 'test')
batch_samples = list()
for im in images:
@ -609,7 +637,7 @@ class BaseDetector(BaseModel):
return batch_samples
def _postprocess(self, batch_pred):
def postprocess(self, batch_pred):
infer_result = {}
if 'bbox' in batch_pred:
bboxes = batch_pred['bbox']
@ -880,108 +908,24 @@ class PicoDet(BaseDetector):
self.fixed_input_shape = image_shape
return self._define_input_spec(image_shape)
def train(self,
num_epochs,
train_dataset,
train_batch_size=64,
eval_dataset=None,
optimizer=None,
save_interval_epochs=1,
log_interval_steps=10,
save_dir='output',
pretrain_weights='IMAGENET',
learning_rate=.001,
warmup_steps=0,
warmup_start_lr=0.0,
lr_decay_epochs=(216, 243),
lr_decay_gamma=0.1,
metric=None,
use_ema=False,
early_stop=False,
early_stop_patience=5,
use_vdl=True,
resume_checkpoint=None):
"""
Train the model.
Args:
num_epochs (int): Number of epochs.
train_dataset (paddlers.datasets.COCODetDataset|paddlers.datasets.VOCDetDataset):
Training dataset.
train_batch_size (int, optional): Total batch size among all cards used in
training. Defaults to 64.
eval_dataset (paddlers.datasets.COCODetDataset|paddlers.datasets.VOCDetDataset|None, optional):
Evaluation dataset. If None, the model will not be evaluated during training
process. Defaults to None.
optimizer (paddle.optimizer.Optimizer|None, optional): Optimizer used for
training. If None, a default optimizer will be used. Defaults to None.
save_interval_epochs (int, optional): Epoch interval for saving the model.
Defaults to 1.
log_interval_steps (int, optional): Step interval for printing training
information. Defaults to 10.
save_dir (str, optional): Directory to save the model. Defaults to 'output'.
pretrain_weights (str|None, optional): None or name/path of pretrained
weights. If None, no pretrained weights will be loaded.
Defaults to 'IMAGENET'.
learning_rate (float, optional): Learning rate for training. Defaults to .001.
warmup_steps (int, optional): Number of steps of warm-up training.
Defaults to 0.
warmup_start_lr (float, optional): Start learning rate of warm-up training.
Defaults to 0..
lr_decay_epochs (list|tuple, optional): Epoch milestones for learning
rate decay. Defaults to (216, 243).
lr_decay_gamma (float, optional): Gamma coefficient of learning rate decay.
Defaults to .1.
metric (str|None, optional): Evaluation metric. Choices are {'VOC', 'COCO', None}.
If None, determine the metric according to the dataset format.
Defaults to None.
use_ema (bool, optional): Whether to use exponential moving average
strategy. Defaults to False.
early_stop (bool, optional): Whether to adopt early stop strategy.
Defaults to False.
early_stop_patience (int, optional): Early stop patience. Defaults to 5.
use_vdl(bool, optional): Whether to use VisualDL to monitor the training
process. Defaults to True.
resume_checkpoint (str|None, optional): Path of the checkpoint to resume
training from. If None, no training checkpoint will be resumed. At most
Aone of `resume_checkpoint` and `pretrain_weights` can be set simultaneously.
Defaults to None.
"""
def _pre_train(self, in_args):
optimizer = in_args['optimizer']
if optimizer is None:
num_steps_each_epoch = len(train_dataset) // train_batch_size
num_steps_each_epoch = len(in_args['train_dataset']) // in_args[
'train_batch_size']
optimizer = self.default_optimizer(
parameters=self.net.parameters(),
learning_rate=learning_rate,
warmup_steps=warmup_steps,
warmup_start_lr=warmup_start_lr,
lr_decay_epochs=lr_decay_epochs,
lr_decay_gamma=lr_decay_gamma,
num_steps_each_epoch=num_steps_each_epoch,
learning_rate=in_args['learning_rate'],
warmup_steps=in_args['warmup_steps'],
warmup_start_lr=in_args['warmup_start_lr'],
lr_decay_epochs=in_args['lr_decay_epochs'],
lr_decay_gamma=in_args['lr_decay_gamma'],
num_steps_each_epoch=in_args['num_steps_each_epoch'],
reg_coeff=4e-05,
scheduler='Cosine',
num_epochs=num_epochs)
super(PicoDet, self).train(
num_epochs=num_epochs,
train_dataset=train_dataset,
train_batch_size=train_batch_size,
eval_dataset=eval_dataset,
optimizer=optimizer,
save_interval_epochs=save_interval_epochs,
log_interval_steps=log_interval_steps,
save_dir=save_dir,
pretrain_weights=pretrain_weights,
learning_rate=learning_rate,
warmup_steps=warmup_steps,
warmup_start_lr=warmup_start_lr,
lr_decay_epochs=lr_decay_epochs,
lr_decay_gamma=lr_decay_gamma,
metric=metric,
use_ema=use_ema,
early_stop=early_stop,
early_stop_patience=early_stop_patience,
use_vdl=use_vdl,
resume_checkpoint=resume_checkpoint)
num_epochs=in_args['num_epochs'])
in_args['optimizer'] = optimizer
return in_args
def build_data_loader(self, dataset, batch_size, mode='train'):
if dataset.num_samples < batch_size:
@ -1393,82 +1337,12 @@ class FasterRCNN(BaseDetector):
super(FasterRCNN, self).__init__(
model_name='FasterRCNN', num_classes=num_classes, **params)
def train(self,
num_epochs,
train_dataset,
train_batch_size=64,
eval_dataset=None,
optimizer=None,
save_interval_epochs=1,
log_interval_steps=10,
save_dir='output',
pretrain_weights='IMAGENET',
learning_rate=.001,
warmup_steps=0,
warmup_start_lr=0.0,
lr_decay_epochs=(216, 243),
lr_decay_gamma=0.1,
metric=None,
use_ema=False,
early_stop=False,
early_stop_patience=5,
use_vdl=True,
resume_checkpoint=None):
"""
Train the model.
Args:
num_epochs (int): Number of epochs.
train_dataset (paddlers.datasets.COCODetDataset|paddlers.datasets.VOCDetDataset):
Training dataset.
train_batch_size (int, optional): Total batch size among all cards used in
training. Defaults to 64.
eval_dataset (paddlers.datasets.COCODetDataset|paddlers.datasets.VOCDetDataset|None, optional):
Evaluation dataset. If None, the model will not be evaluated during training
process. Defaults to None.
optimizer (paddle.optimizer.Optimizer|None, optional): Optimizer used for
training. If None, a default optimizer will be used. Defaults to None.
save_interval_epochs (int, optional): Epoch interval for saving the model.
Defaults to 1.
log_interval_steps (int, optional): Step interval for printing training
information. Defaults to 10.
save_dir (str, optional): Directory to save the model. Defaults to 'output'.
pretrain_weights (str|None, optional): None or name/path of pretrained
weights. If None, no pretrained weights will be loaded.
Defaults to 'IMAGENET'.
learning_rate (float, optional): Learning rate for training. Defaults to .001.
warmup_steps (int, optional): Number of steps of warm-up training.
Defaults to 0.
warmup_start_lr (float, optional): Start learning rate of warm-up training.
Defaults to 0..
lr_decay_epochs (list|tuple, optional): Epoch milestones for learning
rate decay. Defaults to (216, 243).
lr_decay_gamma (float, optional): Gamma coefficient of learning rate decay.
Defaults to .1.
metric (str|None, optional): Evaluation metric. Choices are {'VOC', 'COCO', None}.
If None, determine the metric according to the dataset format.
Defaults to None.
use_ema (bool, optional): Whether to use exponential moving average
strategy. Defaults to False.
early_stop (bool, optional): Whether to adopt early stop strategy.
Defaults to False.
early_stop_patience (int, optional): Early stop patience. Defaults to 5.
use_vdl(bool, optional): Whether to use VisualDL to monitor the training
process. Defaults to True.
resume_checkpoint (str|None, optional): Path of the checkpoint to resume
training from. If None, no training checkpoint will be resumed. At most
Aone of `resume_checkpoint` and `pretrain_weights` can be set simultaneously.
Defaults to None.
"""
def _pre_train(self, in_args):
train_dataset = in_args['train_dataset']
if train_dataset.pos_num < len(train_dataset.file_list):
# In-place modification
train_dataset.num_workers = 0
super(FasterRCNN, self).train(
num_epochs, train_dataset, train_batch_size, eval_dataset,
optimizer, save_interval_epochs, log_interval_steps, save_dir,
pretrain_weights, learning_rate, warmup_steps, warmup_start_lr,
lr_decay_epochs, lr_decay_gamma, metric, use_ema, early_stop,
early_stop_patience, use_vdl, resume_checkpoint)
return in_args
def _compose_batch_transform(self, transforms, mode='train'):
if mode == 'train':
@ -2235,82 +2109,12 @@ class MaskRCNN(BaseDetector):
super(MaskRCNN, self).__init__(
model_name='MaskRCNN', num_classes=num_classes, **params)
def train(self,
num_epochs,
train_dataset,
train_batch_size=64,
eval_dataset=None,
optimizer=None,
save_interval_epochs=1,
log_interval_steps=10,
save_dir='output',
pretrain_weights='IMAGENET',
learning_rate=.001,
warmup_steps=0,
warmup_start_lr=0.0,
lr_decay_epochs=(216, 243),
lr_decay_gamma=0.1,
metric=None,
use_ema=False,
early_stop=False,
early_stop_patience=5,
use_vdl=True,
resume_checkpoint=None):
"""
Train the model.
Args:
num_epochs (int): Number of epochs.
train_dataset (paddlers.datasets.COCODetDataset|paddlers.datasets.VOCDetDataset):
Training dataset.
train_batch_size (int, optional): Total batch size among all cards used in
training. Defaults to 64.
eval_dataset (paddlers.datasets.COCODetDataset|paddlers.datasets.VOCDetDataset|None, optional):
Evaluation dataset. If None, the model will not be evaluated during training
process. Defaults to None.
optimizer (paddle.optimizer.Optimizer|None, optional): Optimizer used for
training. If None, a default optimizer will be used. Defaults to None.
save_interval_epochs (int, optional): Epoch interval for saving the model.
Defaults to 1.
log_interval_steps (int, optional): Step interval for printing training
information. Defaults to 10.
save_dir (str, optional): Directory to save the model. Defaults to 'output'.
pretrain_weights (str|None, optional): None or name/path of pretrained
weights. If None, no pretrained weights will be loaded.
Defaults to 'IMAGENET'.
learning_rate (float, optional): Learning rate for training. Defaults to .001.
warmup_steps (int, optional): Number of steps of warm-up training.
Defaults to 0.
warmup_start_lr (float, optional): Start learning rate of warm-up training.
Defaults to 0..
lr_decay_epochs (list|tuple, optional): Epoch milestones for learning
rate decay. Defaults to (216, 243).
lr_decay_gamma (float, optional): Gamma coefficient of learning rate decay.
Defaults to .1.
metric (str|None, optional): Evaluation metric. Choices are {'VOC', 'COCO', None}.
If None, determine the metric according to the dataset format.
Defaults to None.
use_ema (bool, optional): Whether to use exponential moving average
strategy. Defaults to False.
early_stop (bool, optional): Whether to adopt early stop strategy.
Defaults to False.
early_stop_patience (int, optional): Early stop patience. Defaults to 5.
use_vdl(bool, optional): Whether to use VisualDL to monitor the training
process. Defaults to True.
resume_checkpoint (str|None, optional): Path of the checkpoint to resume
training from. If None, no training checkpoint will be resumed. At most
Aone of `resume_checkpoint` and `pretrain_weights` can be set simultaneously.
Defaults to None.
"""
def _pre_train(self, in_args):
train_dataset = in_args['train_dataset']
if train_dataset.pos_num < len(train_dataset.file_list):
# In-place modification
train_dataset.num_workers = 0
super(MaskRCNN, self).train(
num_epochs, train_dataset, train_batch_size, eval_dataset,
optimizer, save_interval_epochs, log_interval_steps, save_dir,
pretrain_weights, learning_rate, warmup_steps, warmup_start_lr,
lr_decay_epochs, lr_decay_gamma, metric, use_ema, early_stop,
early_stop_patience, use_vdl, resume_checkpoint)
return in_args
def _compose_batch_transform(self, transforms, mode='train'):
if mode == 'train':

20
paddlers/tasks/segmenter.py
Unescape View File

@ -117,10 +117,10 @@ class BaseSegmenter(BaseModel):
if mode == 'test':
origin_shape = inputs[1]
if self.status == 'Infer':
label_map_list, score_map_list = self._postprocess(
label_map_list, score_map_list = self.postprocess(
net_out, origin_shape, transforms=inputs[2])
else:
logit_list = self._postprocess(
logit_list = self.postprocess(
logit, origin_shape, transforms=inputs[2])
label_map_list = []
score_map_list = []
@ -148,7 +148,7 @@ class BaseSegmenter(BaseModel):
raise ValueError("Expected label.ndim == 4 but got {}".format(
label.ndim))
origin_shape = [label.shape[-2:]]
pred = self._postprocess(
pred = self.postprocess(
pred, origin_shape, transforms=inputs[2])[0] # NCHW
intersect_area, pred_area, label_area = ppseg.utils.metrics.calculate_area(
pred, label, self.num_classes)
@ -526,8 +526,8 @@ class BaseSegmenter(BaseModel):
images = [img_file]
else:
images = img_file
batch_im, batch_origin_shape = self._preprocess(images, transforms,
self.model_type)
batch_im, batch_origin_shape = self.preprocess(images, transforms,
self.model_type)
self.net.eval()
data = (batch_im, batch_origin_shape, transforms.transforms)
outputs = self.run(self.net, data, 'test')
@ -631,7 +631,7 @@ class BaseSegmenter(BaseModel):
dst_data = None
print("GeoTiff saved in {}.".format(save_file))
def _preprocess(self, images, transforms, to_tensor=True):
def preprocess(self, images, transforms, to_tensor=True):
self._check_transforms(transforms, 'test')
batch_im = list()
batch_ori_shape = list()
@ -698,7 +698,7 @@ class BaseSegmenter(BaseModel):
batch_restore_list.append(restore_list)
return batch_restore_list
def _postprocess(self, batch_pred, batch_origin_shape, transforms):
def postprocess(self, batch_pred, batch_origin_shape, transforms):
batch_restore_list = BaseSegmenter.get_transforms_shape_info(
batch_origin_shape, transforms)
if isinstance(batch_pred, (tuple, list)) and self.status == 'Infer':
@ -786,7 +786,7 @@ class BaseSegmenter(BaseModel):
class UNet(BaseSegmenter):
def __init__(self,
input_channel=3,
in_channels=3,
num_classes=2,
use_mixed_loss=False,
losses=None,
@ -799,7 +799,7 @@ class UNet(BaseSegmenter):
})
super(UNet, self).__init__(
model_name='UNet',
input_channel=input_channel,
input_channel=in_channels,
num_classes=num_classes,
use_mixed_loss=use_mixed_loss,
losses=losses,
@ -808,7 +808,7 @@ class UNet(BaseSegmenter):
class DeepLabV3P(BaseSegmenter):
def __init__(self,
input_channel=3,
in_channels=3,
num_classes=2,
backbone='ResNet50_vd',
use_mixed_loss=False,

12
test_tipc/infer.py
Unescape View File

@ -141,7 +141,7 @@ class TIPCPredictor(object):
return config
def preprocess(self, images, transforms):
preprocessed_samples = self._model._preprocess(
preprocessed_samples = self._model.preprocess(
images, transforms, to_tensor=False)
if self._model.model_type == 'classifier':
preprocessed_samples = {'image': preprocessed_samples[0]}
@ -167,12 +167,12 @@ class TIPCPredictor(object):
def postprocess(self, net_outputs, topk=1, ori_shape=None, transforms=None):
if self._model.model_type == 'classifier':
true_topk = min(self._model.num_classes, topk)
if self._model._postprocess is None:
if self._model.postprocess is None:
self._model.build_postprocess_from_labels(topk)
# XXX: Convert ndarray to tensor as self._model._postprocess requires
# XXX: Convert ndarray to tensor as self._model.postprocess requires
assert len(net_outputs) == 1
net_outputs = paddle.to_tensor(net_outputs[0])
outputs = self._model._postprocess(net_outputs)
outputs = self._model.postprocess(net_outputs)
class_ids = map(itemgetter('class_ids'), outputs)
scores = map(itemgetter('scores'), outputs)
label_names = map(itemgetter('label_names'), outputs)
@ -182,7 +182,7 @@ class TIPCPredictor(object):
'label_names_map': n,
} for l, s, n in zip(class_ids, scores, label_names)]
elif self._model.model_type in ('segmenter', 'change_detector'):
label_map, score_map = self._model._postprocess(
label_map, score_map = self._model.postprocess(
net_outputs,
batch_origin_shape=ori_shape,
transforms=transforms.transforms)
@ -195,7 +195,7 @@ class TIPCPredictor(object):
k: v
for k, v in zip(['bbox', 'bbox_num', 'mask'], net_outputs)
}
preds = self._model._postprocess(net_outputs)
preds = self._model.postprocess(net_outputs)
else:
logging.error(
"Invalid model type {}.".format(self._model.model_type),

Write Preview
Loading…
Cancel
Save