drop_last=DROP_LAST)
NUM_CLASS = len(train_loader.dataset.classes)
print("Number of Training Classes: {}".format(NUM_CLASS))
BACKBONE = ResNet_50(INPUT_SIZE)
print("=" * 60)
print(BACKBONE)
print("{} Backbone Generated".format(BACKBONE_NAME))
MaskNet = MaskNet()
print("=" * 60)
print(MaskNet)
print("MaskNet Generated")
masknet_paras_only_bn, masknet_paras_wo_bn = separate_irse_bn_paras(
MaskNet
) # separate batch_norm parameters from others; do not do weight decay for batch_norm parameters to improve the generalizability
OPTIMIZER = optim.SGD([{
'params': masknet_paras_wo_bn,
'weight_decay': WEIGHT_DECAY
}, {
'params': masknet_paras_only_bn
}],
lr=LR,
momentum=MOMENTUM)
print("=" * 60)
print(OPTIMIZER)
print("Optimizer Generated")
print("=" * 60)
# optionally resume from a checkpoint
HEAD = HEAD_DICT[HEAD_NAME]
logger.info("=" * 60)
logger.info(HEAD)
logger.info("{} Head Generated".format(HEAD_NAME))
logger.info("=" * 60)
LOSS_DICT = {'Focal': FocalLoss(), 'Softmax': nn.CrossEntropyLoss()}
LOSS = LOSS_DICT[LOSS_NAME]
logger.info("=" * 60)
logger.info(LOSS)
logger.info("{} Loss Generated".format(LOSS_NAME))
logger.info("=" * 60)
if BACKBONE_NAME.find("IR") >= 0:
backbone_paras_only_bn, backbone_paras_wo_bn = separate_irse_bn_paras(
BACKBONE
) # separate batch_norm parameters from others; do not do weight decay for batch_norm parameters to improve the generalizability
_, head_paras_wo_bn = separate_irse_bn_paras(HEAD)
else:
backbone_paras_only_bn, backbone_paras_wo_bn = separate_resnet_bn_paras(
BACKBONE
) # separate batch_norm parameters from others; do not do weight decay for batch_norm parameters to improve the generalizability
_, head_paras_wo_bn = separate_resnet_bn_paras(HEAD)
DISP_FREQ = len(train_loader) # frequency to display training loss & acc
NUM_EPOCH_WARM_UP = NUM_EPOCH // 25 # use the first 1/25 epochs to warm up
NUM_BATCH_WARM_UP = len(
train_loader
) * NUM_EPOCH_WARM_UP # use the first 1/25 epochs to warm up
scheduler = paddle.optimizer.lr.LinearWarmup(
learning_rate=LR,