Skip to content
This repository has been archived by the owner on Oct 31, 2023. It is now read-only.

add color jitter augmentation #680

Merged
merged 2 commits into from
Apr 19, 2019
Merged

add color jitter augmentation #680

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
18bda0a
add color jitter augmentation
CoinCheung Apr 16, 2019
dd5003d
fix spelling
CoinCheung Apr 17, 2019
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
6 changes: 6 additions & 0 deletions maskrcnn_benchmark/config/defaults.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,6 +54,12 @@
# Convert image to BGR format (for Caffe2 models), in range 0-255
_C.INPUT.TO_BGR255 = True

# Image ColorJitter
_C.INPUT.BRIGHTNETSS = 0.0
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

typo: BRIGHTNESS

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hi,

I have correct the spelling, please tell me if there needs further modifications.

As for the performance, I actually observed no obvious improvements on our own datasets with brightness=0.4, contrast=0.4, saturation=0.4, hue=0.3. The random seed is not manually assigned and the performance is on par with the non-color-aug version.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

By the way, do you think we should add a random_seed field in the configuration file for convenience of assigning the seed for random operations?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Having a seed would be useful I think, even though not everything would be deterministic.

_C.INPUT.CONTRAST = 0.0
_C.INPUT.SITURATION = 0.0
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

typo: it should be SATURATION

_C.INPUT.HUE = 0.0


# -----------------------------------------------------------------------------
# Dataset
Expand Down
7 changes: 7 additions & 0 deletions maskrcnn_benchmark/data/transforms/build.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,9 +16,16 @@ def build_transforms(cfg, is_train=True):
normalize_transform = T.Normalize(
mean=cfg.INPUT.PIXEL_MEAN, std=cfg.INPUT.PIXEL_STD, to_bgr255=to_bgr255
)
color_jitter = T.ColorJitter(
brightness=cfg.INPUT.BRIGHTNETSS,
contrast=cfg.INPUT.CONTRAST,
saturation=cfg.INPUT.SITURATION,
hue=cfg.INPUT.HUE,
)

transform = T.Compose(
[
color_jitter,
T.Resize(min_size, max_size),
T.RandomHorizontalFlip(flip_prob),
T.ToTensor(),
Expand Down
18 changes: 18 additions & 0 deletions maskrcnn_benchmark/data/transforms/transforms.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,6 +72,24 @@ def __call__(self, image, target):
return image, target


class ColorJitter(object):
def __init__(self,
brightness=None,
contrast=None,
saturation=None,
hue=None,
):
self.color_jitter = torchvision.transforms.ColorJitter(
brightness=brightness,
contrast=contrast,
saturation=saturation,
hue=hue,)

def __call__(self, image, target):
image = self.color_jitter(image)
return image, target


class ToTensor(object):
def __call__(self, image, target):
return F.to_tensor(image), target
Expand Down