Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Add activation function option for yolov8 detector. #2489

Open
wants to merge 3 commits into
base: master
Choose a base branch
from
Open

Add activation function option for yolov8 detector. #2489

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
8072eb5
Add activation function option for detector.
digital-nomad-cheng Sep 13, 2024
a2baa3a
Merge branch 'keras-team:master' into master
digital-nomad-cheng Sep 18, 2024
82f8e7f
Fix format.
digital-nomad-cheng Sep 18, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
29 changes: 17 additions & 12 deletions keras_cv/src/models/object_detection/yolo_v8/yolo_v8_detector.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -104,7 +104,9 @@ def get_anchors(
return all_anchors, all_strides


def apply_path_aggregation_fpn(features, depth=3, name="fpn"):
def apply_path_aggregation_fpn(
features, depth=3, activation="swish", name="fpn"
):
"""Applies the Feature Pyramid Network (FPN) to the outputs of a backbone.

Args:
Expand All @@ -130,7 +132,7 @@ def apply_path_aggregation_fpn(features, depth=3, name="fpn"):
channels=p4.shape[-1],
depth=depth,
shortcut=False,
activation="swish",
activation=activation,
name=f"{name}_p4p5",
)

Expand All @@ -142,7 +144,7 @@ def apply_path_aggregation_fpn(features, depth=3, name="fpn"):
channels=p3.shape[-1],
depth=depth,
shortcut=False,
activation="swish",
activation=activation,
name=f"{name}_p3p4p5",
)

Expand All @@ -152,15 +154,15 @@ def apply_path_aggregation_fpn(features, depth=3, name="fpn"):
p3p4p5.shape[-1],
kernel_size=3,
strides=2,
activation="swish",
activation=activation,
name=f"{name}_p3p4p5_downsample1",
)
p3p4p5_d1 = ops.concatenate([p3p4p5_d1, p4p5], axis=-1)
p3p4p5_d1 = apply_csp_block(
p3p4p5_d1,
channels=p4p5.shape[-1],
shortcut=False,
activation="swish",
activation=activation,
name=f"{name}_p3p4p5_downsample1_block",
)

Expand All @@ -171,15 +173,15 @@ def apply_path_aggregation_fpn(features, depth=3, name="fpn"):
p3p4p5_d1.shape[-1],
kernel_size=3,
strides=2,
activation="swish",
activation=activation,
name=f"{name}_p3p4p5_downsample2",
)
p3p4p5_d2 = ops.concatenate([p3p4p5_d2, p5], axis=-1)
p3p4p5_d2 = apply_csp_block(
p3p4p5_d2,
channels=p5.shape[-1],
shortcut=False,
activation="swish",
activation=activation,
name=f"{name}_p3p4p5_downsample2_block",
)

Expand All @@ -189,6 +191,7 @@ def apply_path_aggregation_fpn(features, depth=3, name="fpn"):
def apply_yolo_v8_head(
inputs,
num_classes,
activation="swish",
name="yolo_v8_head",
):
"""Applies a YOLOV8 head.
Expand Down Expand Up @@ -229,14 +232,14 @@ def apply_yolo_v8_head(
feature,
box_channels,
kernel_size=3,
activation="swish",
activation=activation,
name=f"{cur_name}_box_1",
)
box_predictions = apply_conv_bn(
box_predictions,
box_channels,
kernel_size=3,
activation="swish",
activation=activation,
name=f"{cur_name}_box_2",
)
box_predictions = keras.layers.Conv2D(
Expand All @@ -249,14 +252,14 @@ def apply_yolo_v8_head(
feature,
class_channels,
kernel_size=3,
activation="swish",
activation=activation,
name=f"{cur_name}_class_1",
)
class_predictions = apply_conv_bn(
class_predictions,
class_channels,
kernel_size=3,
activation="swish",
activation=activation,
name=f"{cur_name}_class_2",
)
class_predictions = keras.layers.Conv2D(
Expand Down Expand Up @@ -400,6 +403,7 @@ def __init__(
num_classes,
bounding_box_format,
fpn_depth=2,
activation="swish",
label_encoder=None,
prediction_decoder=None,
**kwargs,
Expand All @@ -416,12 +420,13 @@ def __init__(
features = list(feature_extractor(images).values())

fpn_features = apply_path_aggregation_fpn(
features, depth=fpn_depth, name="pa_fpn"
features, depth=fpn_depth, activation=activation, name="pa_fpn"
)

outputs = apply_yolo_v8_head(
fpn_features,
num_classes,
activation=activation,
)

# To make loss metrics pretty, we use a no-op layer with a good name.
Expand Down
Loading