def get_estimator(batch_size=4,
epochs=2,
max_train_steps_per_epoch=None,
log_steps=100,
style_weight=5.0,
content_weight=1.0,
tv_weight=1e-4,
save_dir=tempfile.mkdtemp(),
style_img_path='Vassily_Kandinsky,_1913_-_Composition_7.jpg',
data_dir=None):
train_data, _ = mscoco.load_data(root_dir=data_dir, load_bboxes=False, load_masks=False, load_captions=False)
device = "cuda" if torch.cuda.is_available() else "cpu"
style_img = cv2.imread(style_img_path)
assert style_img is not None, "cannot load the style image, please go to the folder with style image"
style_img = cv2.resize(style_img, (256, 256))
style_img = (style_img.astype(np.float32) - 127.5) / 127.5
pipeline = fe.Pipeline(
train_data=train_data,
batch_size=batch_size,
ops=[
ReadImage(inputs="image", outputs="image"),
Normalize(inputs="image", outputs="image", mean=1.0, std=1.0, max_pixel_value=127.5),
Resize(height=256, width=256, image_in="image", image_out="image"),
LambdaOp(fn=lambda: style_img, outputs="style_image"),
ChannelTranspose(inputs=["image", "style_image"], outputs=["image", "style_image"])
])
model = fe.build(model_fn=StyleTransferNet,
model_name="style_transfer_net",
optimizer_fn=lambda x: torch.optim.Adam(x, lr=1e-3))
network = fe.Network(ops=[
ModelOp(inputs="image", model=model, outputs="image_out"),
ExtractVGGFeatures(inputs="style_image", outputs="y_style", device=device),
ExtractVGGFeatures(inputs="image", outputs="y_content", device=device),
ExtractVGGFeatures(inputs="image_out", outputs="y_pred", device=device),
StyleContentLoss(style_weight=style_weight,
content_weight=content_weight,
tv_weight=tv_weight,
inputs=('y_pred', 'y_style', 'y_content', 'image_out'),
outputs='loss'),
UpdateOp(model=model, loss_name="loss")
])
estimator = fe.Estimator(network=network,
pipeline=pipeline,
traces=ModelSaver(model=model, save_dir=save_dir, frequency=1),
epochs=epochs,
max_train_steps_per_epoch=max_train_steps_per_epoch,
log_steps=log_steps)
return estimator
def get_estimator(data_dir=None,
model_dir=tempfile.mkdtemp(),
epochs=200,
batch_size_per_gpu=32,
train_steps_per_epoch=None,
eval_steps_per_epoch=None):
num_device = get_num_devices()
train_ds, val_ds = mscoco.load_data(root_dir=data_dir)
train_ds = PreMosaicDataset(mscoco_ds=train_ds)
batch_size = num_device * batch_size_per_gpu
pipeline = fe.Pipeline(
train_data=train_ds,
eval_data=val_ds,
ops=[
ReadImage(inputs=("image1", "image2", "image3", "image4"),
outputs=("image1", "image2", "image3", "image4"),
mode="train"),
ReadImage(inputs="image", outputs="image", mode="eval"),
LongestMaxSize(max_size=640,
image_in="image1",
bbox_in="bbox1",
bbox_params=BboxParams("coco", min_area=1.0),
mode="train"),
LongestMaxSize(max_size=640,
image_in="image2",
bbox_in="bbox2",
bbox_params=BboxParams("coco", min_area=1.0),
mode="train"),
LongestMaxSize(max_size=640,
image_in="image3",
bbox_in="bbox3",
bbox_params=BboxParams("coco", min_area=1.0),
mode="train"),
LongestMaxSize(max_size=640,
image_in="image4",
bbox_in="bbox4",
bbox_params=BboxParams("coco", min_area=1.0),
mode="train"),
LongestMaxSize(max_size=640,
image_in="image",
bbox_in="bbox",
bbox_params=BboxParams("coco", min_area=1.0),
mode="eval"),
PadIfNeeded(min_height=640,
min_width=640,
image_in="image",
bbox_in="bbox",
bbox_params=BboxParams("coco", min_area=1.0),
mode="eval",
border_mode=cv2.BORDER_CONSTANT,
value=(114, 114, 114)),
CombineMosaic(inputs=("image1", "image2", "image3", "image4",
"bbox1", "bbox2", "bbox3", "bbox4"),
outputs=("image", "bbox"),
mode="train"),
CenterCrop(height=640,
width=640,
image_in="image",
bbox_in="bbox",
bbox_params=BboxParams("coco", min_area=1.0),
mode="train"),
Sometimes(
HorizontalFlip(image_in="image",
bbox_in="bbox",
bbox_params=BboxParams("coco", min_area=1.0),
mode="train")),
HSVAugment(inputs="image", outputs="image", mode="train"),
ToArray(inputs="bbox", outputs="bbox", dtype="float32"),
CategoryID2ClassID(inputs="bbox", outputs="bbox"),
GTBox(inputs="bbox",
outputs=("gt_sbbox", "gt_mbbox", "gt_lbbox"),
image_size=640),
Delete(keys=("image1", "image2", "image3", "image4", "bbox1",
"bbox2", "bbox3", "bbox4", "bbox"),
mode="train"),
Delete(keys="image_id", mode="eval"),
Batch(batch_size=batch_size, pad_value=0)
])
init_lr = 1e-2 / 64 * batch_size
model = fe.build(
lambda: YoloV5(w=640, h=640, c=3),
optimizer_fn=lambda x: torch.optim.SGD(
x, lr=init_lr, momentum=0.937, weight_decay=0.0005, nesterov=True),
mixed_precision=True)
network = fe.Network(ops=[
RescaleTranspose(inputs="image", outputs="image"),
ModelOp(model=model,
inputs="image",
outputs=("pred_s", "pred_m", "pred_l")),
DecodePred(inputs=("pred_s", "pred_m", "pred_l"),
outputs=("pred_s", "pred_m", "pred_l")),
ComputeLoss(inputs=("pred_s", "gt_sbbox"),
outputs=("sbbox_loss", "sconf_loss", "scls_loss")),
ComputeLoss(inputs=("pred_m", "gt_mbbox"),
outputs=("mbbox_loss", "mconf_loss", "mcls_loss")),
ComputeLoss(inputs=("pred_l", "gt_lbbox"),
outputs=("lbbox_loss", "lconf_loss", "lcls_loss")),
Average(inputs=("sbbox_loss", "mbbox_loss", "lbbox_loss"),
outputs="bbox_loss"),
Average(inputs=("sconf_loss", "mconf_loss", "lconf_loss"),
outputs="conf_loss"),
Average(inputs=("scls_loss", "mcls_loss", "lcls_loss"),
outputs="cls_loss"),
Average(inputs=("bbox_loss", "conf_loss", "cls_loss"),
outputs="total_loss"),
PredictBox(width=640,
height=640,
inputs=("pred_s", "pred_m", "pred_l"),
outputs="box_pred",
mode="eval"),
UpdateOp(model=model, loss_name="total_loss")
])
traces = [
MeanAveragePrecision(num_classes=80,
true_key='bbox',
pred_key='box_pred',
mode="eval"),
BestModelSaver(model=model,
save_dir=model_dir,
metric='mAP',
save_best_mode="max")
]
lr_schedule = {
1:
LRScheduler(model=model,
lr_fn=lambda step: lr_schedule_warmup(
step,
train_steps_epoch=np.ceil(len(train_ds) / batch_size),
init_lr=init_lr)),
4:
LRScheduler(model=model,
lr_fn=lambda epoch: cosine_decay(epoch,
cycle_length=epochs - 3,
init_lr=init_lr,
min_lr=init_lr / 100,
start=4))
}
traces.append(EpochScheduler(lr_schedule))
estimator = fe.Estimator(
pipeline=pipeline,
network=network,
epochs=epochs,
traces=traces,
monitor_names=["bbox_loss", "conf_loss", "cls_loss"],
train_steps_per_epoch=train_steps_per_epoch,
eval_steps_per_epoch=eval_steps_per_epoch)
return estimator
def get_estimator(data_dir=None,
model_dir=tempfile.mkdtemp(),
batch_size=16,
epochs=13,
max_train_steps_per_epoch=None,
max_eval_steps_per_epoch=None,
image_size=512,
num_classes=90):
# pipeline
train_ds, eval_ds = mscoco.load_data(root_dir=data_dir)
pipeline = fe.Pipeline(
train_data=train_ds,
eval_data=eval_ds,
batch_size=batch_size,
ops=[
ReadImage(inputs="image", outputs="image"),
LongestMaxSize(image_size,
image_in="image",
image_out="image",
bbox_in="bbox",
bbox_out="bbox",
bbox_params=BboxParams("coco", min_area=1.0)),
PadIfNeeded(
image_size,
image_size,
border_mode=cv2.BORDER_CONSTANT,
image_in="image",
image_out="image",
bbox_in="bbox",
bbox_out="bbox",
bbox_params=BboxParams("coco", min_area=1.0),
),
Sometimes(
HorizontalFlip(mode="train",
image_in="image",
image_out="image",
bbox_in="bbox",
bbox_out="bbox",
bbox_params='coco')),
# normalize from uint8 to [-1, 1]
Normalize(inputs="image",
outputs="image",
mean=1.0,
std=1.0,
max_pixel_value=127.5),
ShiftLabel(inputs="bbox", outputs="bbox"),
AnchorBox(inputs="bbox",
outputs="anchorbox",
width=image_size,
height=image_size),
ChannelTranspose(inputs="image", outputs="image")
],
pad_value=0)
# network
model = fe.build(model_fn=lambda: RetinaNet(num_classes=num_classes),
optimizer_fn=lambda x: torch.optim.SGD(
x, lr=2e-4, momentum=0.9, weight_decay=0.0001))
network = fe.Network(ops=[
ModelOp(model=model, inputs="image", outputs=["cls_pred", "loc_pred"]),
RetinaLoss(inputs=["anchorbox", "cls_pred", "loc_pred"],
outputs=["total_loss", "focal_loss", "l1_loss"]),
UpdateOp(model=model, loss_name="total_loss"),
PredictBox(input_shape=(image_size, image_size, 3),
inputs=["cls_pred", "loc_pred"],
outputs="pred",
mode="eval")
])
# estimator
traces = [
LRScheduler(model=model, lr_fn=lr_fn),
BestModelSaver(model=model,
save_dir=model_dir,
metric='mAP',
save_best_mode="max"),
MeanAveragePrecision(num_classes=num_classes,
true_key='bbox',
pred_key='pred',
mode="eval")
]
estimator = fe.Estimator(
pipeline=pipeline,
network=network,
epochs=epochs,
traces=traces,
max_train_steps_per_epoch=max_train_steps_per_epoch,
max_eval_steps_per_epoch=max_eval_steps_per_epoch,
monitor_names=["l1_loss", "focal_loss"])
return estimator
def get_estimator(data_dir=None,
epochs=12,
batch_size_per_gpu=4,
im_size=1344,
model_dir=tempfile.mkdtemp(),
train_steps_per_epoch=None,
eval_steps_per_epoch=None):
assert im_size % 32 == 0, "im_size must be a multiple of 32"
num_device = get_num_devices()
train_ds, val_ds = mscoco.load_data(root_dir=data_dir, load_masks=True)
batch_size = num_device * batch_size_per_gpu
pipeline = fe.Pipeline(
train_data=train_ds,
eval_data=val_ds,
test_data=val_ds,
ops=[
ReadImage(inputs="image", outputs="image"),
MergeMask(inputs="mask", outputs="mask"),
GetImageSize(inputs="image", outputs="imsize", mode="test"),
LongestMaxSize(max_size=im_size,
image_in="image",
mask_in="mask",
bbox_in="bbox",
bbox_params="coco"),
RemoveIf(fn=lambda x: len(x) == 0, inputs="bbox"),
PadIfNeeded(min_height=im_size,
min_width=im_size,
image_in="image",
mask_in="mask",
bbox_in="bbox",
bbox_params="coco",
border_mode=cv2.BORDER_CONSTANT,
value=0),
Sometimes(
HorizontalFlip(image_in="image",
mask_in="mask",
bbox_in="bbox",
bbox_params="coco",
mode="train")),
Resize(height=im_size // 4, width=im_size // 4,
image_in='mask'), # downscale mask for memory efficiency
Gt2Target(inputs=("mask", "bbox"),
outputs=("gt_match", "mask", "classes")),
Delete(keys="bbox"),
Delete(keys="image_id", mode="!test"),
Batch(batch_size=batch_size, pad_value=0)
],
num_process=8 * num_device)
init_lr = 1e-2 / 16 * batch_size
model = fe.build(
model_fn=SoloV2,
optimizer_fn=lambda x: torch.optim.SGD(x, lr=init_lr, momentum=0.9))
network = fe.Network(ops=[
Normalize(inputs="image",
outputs="image",
mean=(0.485, 0.456, 0.406),
std=(0.229, 0.224, 0.225)),
Permute(inputs="image", outputs='image'),
ModelOp(model=model,
inputs="image",
outputs=("feat_seg", "feat_cls_list", "feat_kernel_list")),
LambdaOp(fn=lambda x: x,
inputs="feat_cls_list",
outputs=("cls1", "cls2", "cls3", "cls4", "cls5")),
LambdaOp(fn=lambda x: x,
inputs="feat_kernel_list",
outputs=("k1", "k2", "k3", "k4", "k5")),
Solov2Loss(0,
40,
inputs=("mask", "classes", "gt_match", "feat_seg", "cls1",
"k1"),
outputs=("l_c1", "l_s1")),
Solov2Loss(1,
36,
inputs=("mask", "classes", "gt_match", "feat_seg", "cls2",
"k2"),
outputs=("l_c2", "l_s2")),
Solov2Loss(2,
24,
inputs=("mask", "classes", "gt_match", "feat_seg", "cls3",
"k3"),
outputs=("l_c3", "l_s3")),
Solov2Loss(3,
16,
inputs=("mask", "classes", "gt_match", "feat_seg", "cls4",
"k4"),
outputs=("l_c4", "l_s4")),
Solov2Loss(4,
12,
inputs=("mask", "classes", "gt_match", "feat_seg", "cls5",
"k5"),
outputs=("l_c5", "l_s5")),
CombineLoss(inputs=("l_c1", "l_s1", "l_c2", "l_s2", "l_c3", "l_s3",
"l_c4", "l_s4", "l_c5", "l_s5"),
outputs=("total_loss", "cls_loss", "seg_loss")),
L2Regularizaton(inputs="total_loss",
outputs="total_loss_l2",
model=model,
beta=1e-5,
mode="train"),
UpdateOp(model=model, loss_name="total_loss_l2"),
PointsNMS(inputs="feat_cls_list", outputs="feat_cls_list",
mode="test"),
Predict(inputs=("feat_seg", "feat_cls_list", "feat_kernel_list"),
outputs=("seg_preds", "cate_scores", "cate_labels"),
mode="test")
])
train_steps_epoch = int(np.ceil(len(train_ds) / batch_size))
lr_schedule = {
1:
LRScheduler(
model=model,
lr_fn=lambda step: lr_schedule_warmup(step, init_lr=init_lr)),
2:
LRScheduler(
model=model,
lr_fn=lambda step: cosine_decay(step,
cycle_length=train_steps_epoch *
(epochs - 1),
init_lr=init_lr,
min_lr=init_lr / 100,
start=train_steps_epoch))
}
traces = [
EpochScheduler(lr_schedule),
COCOMaskmAP(data_dir=val_ds.root_dir,
inputs=("seg_preds", "cate_scores", "cate_labels",
"image_id", "imsize"),
mode="test"),
BestModelSaver(model=model, save_dir=model_dir, metric="total_loss")
]
estimator = fe.Estimator(pipeline=pipeline,
network=network,
epochs=epochs,
traces=traces,
monitor_names=("cls_loss", "seg_loss"),
train_steps_per_epoch=train_steps_per_epoch,
eval_steps_per_epoch=eval_steps_per_epoch)
return estimator
