refine code and docs
parent
1c31010b14
commit
790815f430
ppcls
|
@ -33,8 +33,8 @@ def get_architectures():
|
|||
|
||||
|
||||
def get_blacklist_model_in_static_mode():
|
||||
from ppcls.arch.backbone import (distilled_vision_transformer,
|
||||
vision_transformer)
|
||||
from ppcls.arch.backbone import distilled_vision_transformer
|
||||
from ppcls.arch.backbone import vision_transformer
|
||||
blacklist = distilled_vision_transformer.__all__ + vision_transformer.__all__
|
||||
return blacklist
|
||||
|
||||
|
@ -60,10 +60,10 @@ def get_param_attr_dict(ParamAttr_config: Union[None, bool, Dict[str, Dict]]
|
|||
"""parse ParamAttr from an dict
|
||||
|
||||
Args:
|
||||
ParamAttr_config (Union[bool, Dict[str, Dict]]): ParamAttr_config
|
||||
ParamAttr_config (Union[None, bool, Dict[str, Dict]]): ParamAttr configure
|
||||
|
||||
Returns:
|
||||
Union[bool, paddle.ParamAttr]: Generated ParamAttr
|
||||
Union[None, bool, paddle.ParamAttr]: Generated ParamAttr
|
||||
"""
|
||||
if ParamAttr_config is None:
|
||||
return None
|
||||
|
|
|
@ -22,7 +22,6 @@ import six
|
|||
import math
|
||||
import random
|
||||
import cv2
|
||||
from typing import Sequence
|
||||
import numpy as np
|
||||
from PIL import Image, ImageOps, __version__ as PILLOW_VERSION
|
||||
from paddle.vision.transforms import ColorJitter as RawColorJitter
|
||||
|
|
|
@ -23,8 +23,9 @@ import paddle.nn as nn
|
|||
|
||||
|
||||
class CenterLoss(nn.Layer):
|
||||
"""Center loss class
|
||||
|
||||
"""Center loss
|
||||
paper : [A Discriminative Feature Learning Approach for Deep Face Recognition](https://link.springer.com/content/pdf/10.1007%2F978-3-319-46478-7_31.pdf)
|
||||
code reference: https://github.com/michuanhaohao/reid-strong-baseline/blob/master/layers/center_loss.py#L7
|
||||
Args:
|
||||
num_classes (int): number of classes.
|
||||
feat_dim (int): number of feature dimensions.
|
||||
|
|
|
@ -71,7 +71,7 @@ def build_optimizer(config, epochs, step_each_epoch, model_list=None):
|
|||
optim_cfg = optim_item[optim_name] # get optim_cfg
|
||||
|
||||
lr = build_lr_scheduler(optim_cfg.pop('lr'), epochs, step_each_epoch)
|
||||
logger.info("build lr ({}) for scope ({}) success..".format(
|
||||
logger.debug("build lr ({}) for scope ({}) success..".format(
|
||||
lr, optim_scope))
|
||||
# step2 build regularization
|
||||
if 'regularizer' in optim_cfg and optim_cfg['regularizer'] is not None:
|
||||
|
@ -83,8 +83,8 @@ def build_optimizer(config, epochs, step_each_epoch, model_list=None):
|
|||
reg_name = reg_config.pop('name') + 'Decay'
|
||||
reg = getattr(paddle.regularizer, reg_name)(**reg_config)
|
||||
optim_cfg["weight_decay"] = reg
|
||||
logger.info("build regularizer ({}) for scope ({}) success..".
|
||||
format(reg, optim_scope))
|
||||
logger.debug("build regularizer ({}) for scope ({}) success..".
|
||||
format(reg, optim_scope))
|
||||
# step3 build optimizer
|
||||
if 'clip_norm' in optim_cfg:
|
||||
clip_norm = optim_cfg.pop('clip_norm')
|
||||
|
@ -123,7 +123,7 @@ def build_optimizer(config, epochs, step_each_epoch, model_list=None):
|
|||
optim = getattr(optimizer, optim_name)(
|
||||
learning_rate=lr, grad_clip=grad_clip,
|
||||
**optim_cfg)(model_list=optim_model)
|
||||
logger.info("build optimizer ({}) for scope ({}) success..".format(
|
||||
logger.debug("build optimizer ({}) for scope ({}) success..".format(
|
||||
optim, optim_scope))
|
||||
optim_list.append(optim)
|
||||
lr_list.append(lr)
|
||||
|
|
|
@ -262,24 +262,6 @@ class Piecewise(object):
|
|||
return learning_rate
|
||||
|
||||
|
||||
class Constant(LRScheduler):
|
||||
"""
|
||||
Constant learning rate
|
||||
Args:
|
||||
lr (float): The initial learning rate. It is a python float number.
|
||||
last_epoch (int, optional): The index of last epoch. Can be set to restart training. Default: -1, means initial learning rate.
|
||||
"""
|
||||
|
||||
def __init__(self, learning_rate, last_epoch=-1, by_epoch=False, **kwargs):
|
||||
self.learning_rate = learning_rate
|
||||
self.last_epoch = last_epoch
|
||||
self.by_epoch = by_epoch
|
||||
super().__init__()
|
||||
|
||||
def get_lr(self):
|
||||
return self.learning_rate
|
||||
|
||||
|
||||
class MultiStepDecay(LRScheduler):
|
||||
"""
|
||||
Update the learning rate by ``gamma`` once ``epoch`` reaches one of the milestones.
|
||||
|
|
Loading…
Reference in New Issue