batch_size, eval_batch_size = 4 * mul, 4
ds_train, ds_val, steps_per_epoch, val_steps = make_voc_dataset_sub(
n_train, n_val, batch_size, eval_batch_size, preprocess)
backbone = resnet10()
model = FCOS(backbone,
num_classes=20,
feat_channels=64,
stacked_convs=2,
norm='bn')
model.build((None, HEIGHT, WIDTH, 3))
# load_checkpoint("./drive/MyDrive/models/ImageNet-86/ckpt", model=backbone)
criterion = DetectionLoss(box_loss_fn=iou_loss(mode='ciou', offset=True),
cls_loss_fn=focal_loss(alpha=0.25, gamma=2.0),
centerness=True)
base_lr = 0.0025
epochs = 60
lr_schedule = CosineLR(base_lr * mul,
steps_per_epoch,
epochs,
min_lr=0,
warmup_min_lr=base_lr,
warmup_epoch=5)
optimizer = SGD(lr_schedule, momentum=0.9, nesterov=True, weight_decay=1e-4)
train_metrics = {
'loss': Mean(),
}
eval_metrics = {
return image, {'bbox_target': bbox_targets, 'label': labels, 'centerness': centerness,
'image_id': image_id}
mul = 1
n_train, n_val = 8, 8
batch_size, eval_batch_size = 4 * mul, 8
ds_train, ds_val, steps_per_epoch, val_steps = make_dataset_sub(
n_train, n_val, batch_size, eval_batch_size, transform)
backbone = resnet10()
model = ATSS(backbone, num_classes=80, feat_channels=64, stacked_convs=2)
model.build((None, HEIGHT, WIDTH, 3))
criterion = DetectionLoss(
box_loss_fn=iou_loss(mode='giou'), cls_loss_fn=focal_loss(alpha=0.25, gamma=2.0),
bbox_coder=bbox_coder, decode_pred=True, centerness=True)
base_lr = 0.0025
epochs = 100
lr_schedule = CosineLR(base_lr * mul, steps_per_epoch, epochs, min_lr=0,
warmup_min_lr=0, warmup_epoch=5)
optimizer = SGD(lr_schedule, momentum=0.9, nesterov=True, weight_decay=1e-4)
train_metrics = {
'loss': Mean(),
}
eval_metrics = {
'loss': MeanMetricWrapper(criterion),
}