def tta_fn(image, rois):
image = image.permute(1, 2, 0).to('cpu').numpy()
dtype = image.dtype
image = image.astype(np.uint8)
out_images, out_rois = [], []
for tfm_gen in size_gens:
resized_image, tfm = T.apply_transform_gens([tfm_gen], image)
resized_rois = tfm.transforms[0].apply_box(rois.to('cpu').numpy())
if cfg.TEST.AUG.FLIP:
flipped_image, tfm = T.apply_transform_gens([flip], resized_image)
flipped_rois = tfm.transforms[0].apply_box(resized_rois)
img_batch = torch.stack([
torch.from_numpy(resized_image.astype(dtype)).permute(2,0,1),
torch.from_numpy(flipped_image.astype(dtype)).permute(2,0,1)
])
roi_batch = [
torch.from_numpy(resized_rois),
torch.from_numpy(flipped_rois)
]
else:
img_batch = torch.from_numpy(resized_image.astype(dtype)).permute(2,0,1).unsqueeze(0)
roi_batch = [torch.from_numpy(resized_rois),]
out_images.append(img_batch)
out_rois.append(roi_batch)
return out_images, out_rois
def __call__(self, dataset_dict):
"""
Args:
dataset_dict (dict): Metadata of one image, in Detectron2 Dataset format.
Returns:
dict: a format that builtin models in detectron2 accept
"""
dataset_dict = copy.deepcopy(
dataset_dict) # it will be modified by code below
# USER: Write your own image loading if it's not from a file
image = utils.read_image(dataset_dict["file_name"],
format=self.img_format)
utils.check_image_size(dataset_dict, image)
if "annotations" not in dataset_dict:
image, transforms = T.apply_transform_gens(
([self.crop_gen] if self.crop_gen else []) + self.tfm_gens,
image)
else:
# Crop around an instance if there are instances in the image.
# USER: Remove if you don't use cropping
if self.crop_gen:
crop_tfm = utils.gen_crop_transform_with_instance(
self.crop_gen.get_crop_size(image.shape[:2]),
image.shape[:2],
np.random.choice(dataset_dict["annotations"]),
)
image = crop_tfm.apply_image(image)
image, transforms = T.apply_transform_gens(self.tfm_gens, image)
if self.crop_gen:
transforms = crop_tfm + transforms
image_shape = image.shape[:2] # h, w
# Pytorch's dataloader is efficient on torch.Tensor due to shared-memory,
# but not efficient on large generic data structures due to the use of pickle & mp.Queue.
# Therefore it's important to use torch.Tensor.
dataset_dict["image"] = torch.as_tensor(
np.ascontiguousarray(image.transpose(2, 0, 1)))
if not self.is_train:
# USER: Modify this if you want to keep them for some reason.
dataset_dict.pop("annotations", None)
return dataset_dict
if "annotations" in dataset_dict:
# USER: Implement additional transformations if you have other types of data
annos = [
utils.transform_instance_annotations(
obj,
transforms,
image_shape,
) for obj in dataset_dict.pop("annotations")
if obj.get("iscrowd", 0) == 0
]
instances = annotations_to_instances(annos, image_shape)
dataset_dict["instances"] = utils.filter_empty_instances(instances)
return dataset_dict
def __call__(self, dataset_dict):
dataset_dict = copy.deepcopy(dataset_dict)
try:
image = utils.read_image(dataset_dict["file_name"],
format=self.img_format)
utils.check_image_size(dataset_dict, image)
except OSError:
return
if "annotations" not in dataset_dict:
image, transforms = T.apply_transform_gens(
([self.crop_gen] if self.crop_gen else []) + self.tfm_gens,
image)
else:
if self.crop_gen:
crop_tfm = utils.gen_crop_transform_with_instance(
self.crop_gen.get_crop_size(image.shape[:2]),
image.shape[:2],
np.random.choice(dataset_dict["annotations"]),
)
image = crop_tfm.apply_image(image)
image, transforms = T.apply_transform_gens(self.tfm_gens, image)
if self.crop_gen:
transforms = crop_tfm + transforms
image_shape = image.shape[:2]
dataset_dict["image"] = torch.as_tensor(
np.ascontiguousarray(image.transpose(2, 0, 1)))
return dataset_dict
def __call__(self, dataset_dict):
"""
Args:
dataset_dict (dict): Metadata of one image, in Detectron2 Dataset format.
Returns:
dict: a format that builtin models in detectron2 accept
"""
dataset_dict = copy.deepcopy(
dataset_dict) # it will be modified by code below
image = utils.read_image(dataset_dict["file_name"],
format=self.img_format)
utils.check_image_size(dataset_dict, image)
if self.crop_gen is None:
image, transforms = T.apply_transform_gens(self.tfm_gens, image)
else:
if np.random.rand() > 0.5:
image, transforms = T.apply_transform_gens(
self.tfm_gens, image)
else:
image, transforms = T.apply_transform_gens(
self.tfm_gens[:-1] + self.crop_gen + self.tfm_gens[-1:],
image)
image_shape = image.shape[:2] # h, w
# Pytorch's dataloader is efficient on torch.Tensor due to shared-memory,
# but not efficient on large generic data structures due to the use of pickle & mp.Queue.
# Therefore it's important to use torch.Tensor.
dataset_dict["image"] = torch.as_tensor(
np.ascontiguousarray(image.transpose(2, 0, 1)))
if not self.is_train:
# USER: Modify this if you want to keep them for some reason.
dataset_dict.pop("annotations", None)
return dataset_dict
if "annotations" in dataset_dict:
# USER: Modify this if you want to keep them for some reason.
for anno in dataset_dict["annotations"]:
if not self.mask_on:
anno.pop("segmentation", None)
anno.pop("keypoints", None)
# USER: Implement additional transformations if you have other types of data
annos = [
utils.transform_instance_annotations(obj, transforms,
image_shape)
for obj in dataset_dict.pop("annotations")
if obj.get("iscrowd", 0) == 0
]
instances = utils.annotations_to_instances(annos, image_shape)
dataset_dict["instances"] = utils.filter_empty_instances(instances)
return dataset_dict
def test_build_transform_gen(self):
default_cfg = Detectron2GoRunner().get_default_cfg()
default_cfg.INPUT.MIN_SIZE_TRAIN = (30, )
default_cfg.INPUT.MIN_SIZE_TEST = 30
trans_train = build_transform_gen(default_cfg, is_train=True)
trans_test = build_transform_gen(default_cfg, is_train=False)
img = np.zeros((80, 60, 3))
trans_img_train, tl_train = apply_transform_gens(trans_train, img)
trans_img_test, tl_test = apply_transform_gens(trans_test, img)
self.assertEqual(trans_img_train.shape, (40, 30, 3))
self.assertEqual(trans_img_test.shape, (40, 30, 3))
def test_build_transform_gen_resize_square(self):
default_cfg = Detectron2GoRunner().get_default_cfg()
default_cfg.INPUT.MIN_SIZE_TRAIN = (30, )
default_cfg.INPUT.MIN_SIZE_TEST = 40
default_cfg.D2GO_DATA.AUG_OPS.TRAIN = ["ResizeShortestEdgeSquareOp"]
default_cfg.D2GO_DATA.AUG_OPS.TEST = ["ResizeShortestEdgeSquareOp"]
trans_train = build_transform_gen(default_cfg, is_train=True)
trans_test = build_transform_gen(default_cfg, is_train=False)
img = np.zeros((80, 60, 3))
trans_img_train, tl_train = apply_transform_gens(trans_train, img)
trans_img_test, tl_test = apply_transform_gens(trans_test, img)
self.assertEqual(trans_img_train.shape, (30, 30, 3))
self.assertEqual(trans_img_test.shape, (40, 40, 3))
def test_apply_rotated_boxes_unequal_scaling_factor(self):
np.random.seed(125)
h, w = 400, 200
newh, neww = 800, 800
image = np.random.rand(h, w)
transform_gen = []
transform_gen.append(T.Resize(shape=(newh, neww)))
image, transforms = T.apply_transform_gens(transform_gen, image)
image_shape = image.shape[:2] # h, w
assert image_shape == (newh, neww)
boxes = np.array(
[
[150, 100, 40, 20, 0],
[150, 100, 40, 20, 30],
[150, 100, 40, 20, 90],
[150, 100, 40, 20, -90],
],
dtype=np.float64,
)
transformed_boxes = transforms.apply_rotated_box(boxes)
expected_bboxes = np.array(
[
[600, 200, 160, 40, 0],
[600, 200, 144.22205102, 52.91502622, 49.10660535],
[600, 200, 80, 80, 90],
[600, 200, 80, 80, -90],
],
dtype=np.float64,
)
err_msg = "transformed_boxes = {}, expected {}".format(transformed_boxes, expected_bboxes)
assert np.allclose(transformed_boxes, expected_bboxes), err_msg
def __call__(self, dataset_dict):
"""
Args:
dataset_dict (dict): Metadata of one image, in Detectron2 Dataset format.
Returns:
dict: a format that builtin models in detectron2 accept (classification only)
"""
dataset_dict = copy.deepcopy(
dataset_dict) # it will be modified by code below
# USER: Write your own image loading if it's not from a file
image = utils.read_image(dataset_dict["file_name"],
format=self.img_format)
utils.check_image_size(dataset_dict, image)
image, transforms = T.apply_transform_gens(
([self.crop_gen] if self.crop_gen else []) + self.tfm_gens, image)
image_shape = image.shape[:2] # h, w
# Pytorch's dataloader is efficient on torch.Tensor due to shared-memory,
# but not efficient on large generic data structures due to the use of pickle & mp.Queue.
# Therefore it's important to use torch.Tensor.
dataset_dict["image"] = torch.as_tensor(
np.ascontiguousarray(image.transpose(2, 0, 1)))
return dataset_dict
def custom_mapper(dataset_dict):
# it will be modified by code below
dataset_dict = copy.deepcopy(dataset_dict)
image = utils.read_image(dataset_dict["file_name"], format="BGR")
transform_list = [
T.Resize((512, 512)),
T.RandomBrightness(0.8, 1.8),
T.RandomContrast(0.6, 1.3),
T.RandomSaturation(0.8, 1.4),
T.RandomRotation(angle=[30, 30]),
T.RandomLighting(0.7),
T.RandomFlip(prob=0.4, horizontal=False, vertical=True),
]
image, transforms = T.apply_transform_gens(transform_list, image)
dataset_dict["image"] = torch.as_tensor(
image.transpose(2, 0, 1).astype("float32"))
annos = [
utils.transform_instance_annotations(obj, transforms, image.shape[:2])
for obj in dataset_dict.pop("annotations")
if obj.get("iscrowd", 0) == 0
]
instances = utils.annotations_to_instances(annos, image.shape[:2])
dataset_dict["instances"] = utils.filter_empty_instances(instances)
return dataset_dict
def custom_mapper(dataset_dict, size, flip_prob, min_brightness, max_brightness, \
min_contrast, max_contrast, min_saturation, max_saturation):
# Implement a mapper, similar to the default DatasetMapper, but with your own customizations
dataset_dict = copy.deepcopy(dataset_dict) # it will be modified by code below
image = detection_utils.read_image(dataset_dict["file_name"], format="BGR")
transform_list = [
T.Resize(size),
T.RandomBrightness(min_brightness, max_brightness),
T.RandomContrast(min_contrast, max_contrast),
T.RandomSaturation(min_saturation, max_saturation),
T.RandomFlip(prob=flip_prob, horizontal=False, vertical=True),
T.RandomFlip(prob=flip_prob, horizontal=True, vertical=False),
]
image, transforms = T.apply_transform_gens(transform_list, image)
dataset_dict["image"] = torch.as_tensor(image.transpose(2, 0, 1).astype("float32"))
annos = [
detection_utils.transform_instance_annotations(obj, transforms, image.shape[:2])
for obj in dataset_dict.pop("annotations")
if obj.get("iscrowd", 0) == 0
]
instances = detection_utils.annotations_to_instances(annos, image.shape[:2])
dataset_dict["instances"] = detection_utils.filter_empty_instances(instances)
return dataset_dict
def mapper(dataset_dict):
# Implement a mapper, similar to the default DatasetMapper, but with your own customizations
dataset_dict = copy.deepcopy(
dataset_dict) # it will be modified by code below
image = utils.read_image(dataset_dict["file_name"], format="BGR")
image, transforms = T.apply_transform_gens([
T.RandomFlip(prob=0.50, horizontal=True, vertical=False),
T.RandomApply(tfm_or_aug=T.RandomBrightness(intensity_min=0.7,
intensity_max=1.1),
prob=0.40),
T.RandomApply(tfm_or_aug=T.RandomSaturation(intensity_min=0.7,
intensity_max=1.1),
prob=0.40)
], image)
dataset_dict["image"] = torch.as_tensor(
image.transpose(2, 0, 1).astype("float32"))
annos = [
utils.transform_instance_annotations(obj, transforms, image.shape[:2])
for obj in dataset_dict.pop("annotations")
if obj.get("iscrowd", 0) == 0
]
instances = utils.annotations_to_instances(annos, image.shape[:2])
dataset_dict["instances"] = utils.filter_empty_instances(instances)
return dataset_dict
def custom_mapper(dataset_dict):
# Implement a mapper, similar to the default DatasetMapper, but with your own customizations
dataset_dict = copy.deepcopy(
dataset_dict) # it will be modified by code below
image = utils.read_image(dataset_dict["file_name"], format="BGR")
# transform_list = [T.Resize(800,600),
# T.RandomFlip(prob=0.5, horizontal=True, vertical=True),
# T.RandomContrast(0.8, 3),
# T.RandomBrightness(0.8, 1.6),
# ]
transform_list = [ #T.Resize((800, 800)),
T.RandomContrast(0.8, 3),
T.RandomBrightness(0.8, 1.6),
T.RandomFlip(prob=0.5, horizontal=False, vertical=True),
T.RandomFlip(prob=0.5, horizontal=True, vertical=False)
] ### 数据增强方式
image, transforms = T.apply_transform_gens(transform_list,
image) ## # 数组增强
dataset_dict["image"] = torch.as_tensor(
image.transpose(2, 0, 1).astype("float32")) ##转成Tensor
annos = [
utils.transform_instance_annotations(obj, transforms, image.shape[:2])
for obj in dataset_dict.pop("annotations")
if obj.get("iscrowd", 0) == 0
]
instances = utils.annotations_to_instances(
annos, image.shape[:2]) # 将标注转成Instance（Tensor）
dataset_dict["instances"] = utils.filter_empty_instances(
instances) ## 去除空的
return dataset_dict
def mapper(dataset_dict):
# 自定义mapper
dataset_dict = copy.deepcopy(dataset_dict) # 后面要改变这个dict，所以先复制
image = utils.read_image(dataset_dict["file_name"], format="BGR") # 读取图片，numpy array
# image, transforms = T.apply_transform_gens(
# [T.Resize((800, 800)), T.RandomContrast(0.1, 3), T.RandomSaturation(0.1, 2), T.RandomRotation(angle=[0, 180]),
# T.RandomFlip(prob=0.4, horizontal=False, vertical=True), T.RandomCrop('relative_range', (0.4, 0.6))], image) # 数组增强
# image, transforms = T.apply_transform_gens(
# [T.Resize((800, 800)), T.RandomContrast(0.1, 3), T.RandomSaturation(0.1, 2),
# T.RandomFlip(prob=0.4, horizontal=True, vertical=False), T.RandomCrop('relative_range', (0.4, 0.6))], image)
image, transforms = T.apply_transform_gens(
[T.Resize((800, 800)), T.RandomContrast(0.1, 3), T.RandomSaturation(0.1, 2),
T.RandomFlip(prob=0.4, horizontal=True, vertical=False)], image)
# 数组增强
dataset_dict["image"] = torch.as_tensor(image.transpose(2, 0, 1).astype("float32")) # 转成Tensor
annos = [
utils.transform_instance_annotations(obj, transforms, image.shape[:2])
for obj in dataset_dict.pop("annotations")
if obj.get("iscrowd", 0) == 0
] # 数据增强要同步标注
instances = utils.annotations_to_instances(annos, image.shape[:2]) # 将标注转成Instance（Tensor）
dataset_dict["instances"] = utils.filter_empty_instances(instances) # 去除空的
return dataset_dict
def __call__(self, dataset_dict):
"""
Transform the dataset_dict according to the configured transformations.
Args:
dataset_dict (dict): Metadata of one image, in Detectron2 Dataset format.
Returns:
dict: a new dict that's going to be processed by the model.
It currently does the following:
1. Read the image from "file_name"
2. Transform the image and annotations
3. Prepare the annotations to :class:`Instances`
"""
# get 3D models for each annotations and remove 3D mesh models from image dict
mesh_models = []
if "annotations" in dataset_dict:
for anno in dataset_dict["annotations"]:
mesh_models.append(
[
self._all_mesh_models[anno["mesh"]][0].clone(),
self._all_mesh_models[anno["mesh"]][1].clone(),
]
)
dataset_dict = {key: value for key, value in dataset_dict.items() if key != "mesh_models"}
dataset_dict = copy.deepcopy(dataset_dict) # it will be modified by code below
if "annotations" in dataset_dict:
for i, anno in enumerate(dataset_dict["annotations"]):
anno["mesh"] = mesh_models[i]
image = utils.read_image(dataset_dict["file_name"], format=self.img_format)
utils.check_image_size(dataset_dict, image)
image, transforms = T.apply_transform_gens(self.tfm_gens, image)
image_shape = image.shape[:2] # h, w
# Pytorch's dataloader is efficient on torch.Tensor due to shared-memory,
# but not efficient on large generic data structures due to the use of pickle & mp.Queue.
# Therefore it's important to use torch.Tensor.
dataset_dict["image"] = torch.as_tensor(image.transpose(2, 0, 1).astype("float32"))
# Can use uint8 if it turns out to be slow some day
if not self.is_train:
dataset_dict.pop("annotations", None)
return dataset_dict
if "annotations" in dataset_dict:
annos = [
self.transform_annotations(obj, transforms, image_shape)
for obj in dataset_dict.pop("annotations")
if obj.get("iscrowd", 0) == 0
]
# Should not be empty during training
instances = annotations_to_instances(annos, image_shape)
dataset_dict["instances"] = instances[instances.gt_boxes.nonempty()]
return dataset_dict
def __call__(self, dataset_dict):
self.tfm_gens = []
dataset_dict = deepcopy(dataset_dict)
image = utils.read_image(dataset_dict["file_name"],
format=self.img_format)
utils.check_image_size(dataset_dict, image)
if self.is_train:
# Crop
if 'crop' in self.da.keys():
crop_gen = T.RandomCrop(self.da['crop']['type'],
self.da['crop']['size'])
self.tfm_gens.append(crop_gen)
# Horizontal flip
if 'flip' in self.da.keys():
flip_gen = T.RandomFlip(
prob=self.da['flip']['prob'],
horizontal=self.da['flip']['horizontal'],
vertical=self.da['flip']['vertical'])
self.tfm_gens.append(flip_gen)
image, transforms = T.apply_transform_gens(self.tfm_gens, image)
image_shape = image.shape[:2] # h, w
dataset_dict["image"] = torch.as_tensor(
np.ascontiguousarray(image.transpose(2, 0, 1)))
if not self.is_train:
dataset_dict.pop("annotations", None)
dataset_dict.pop("sem_seg_file_name", None)
return dataset_dict
if "annotations" in dataset_dict:
for anno in dataset_dict["annotations"]:
if not self.mask_on:
anno.pop("segmentation", None)
if not self.keypoint_on:
anno.pop("keypoints", None)
annos = [
utils.transform_instance_annotations(
obj,
transforms,
image_shape,
keypoint_hflip_indices=self.keypoint_hflip_indices)
for obj in dataset_dict.pop("annotations")
if obj.get("iscrowd", 0) == 0
]
instances = utils.annotations_to_instances(
annos, image_shape, mask_format=self.mask_format)
if self.crop_gen and instances.has("gt_masks"):
instances.gt_boxes = instances.gt_masks.get_bounding_boxes()
dataset_dict["instances"] = utils.filter_empty_instances(instances)
return dataset_dict
def __call__(self, dataset_dict):
"""
Args:
dataset_dict (dict): Metadata of one image, in Detectron2 Dataset format.
Returns:
dict: a format that builtin models in detectron2 accept
"""
dataset_dict = copy.deepcopy(
dataset_dict) # it will be modified by code below
image = utils.read_image(dataset_dict["file_name"],
format=self.img_format)
utils.check_image_size(dataset_dict, image)
image, transforms = T.apply_transform_gens(self.augmentation, image)
image_shape = image.shape[:2] # h, w
dataset_dict["image"] = torch.as_tensor(
image.transpose(2, 0, 1).astype("float32"))
if not self.is_train:
dataset_dict.pop("annotations", None)
return dataset_dict
for anno in dataset_dict["annotations"]:
if not self.mask_on:
anno.pop("segmentation", None)
if not self.keypoint_on:
anno.pop("keypoints", None)
# USER: Implement additional transformations if you have other types of data
# USER: Don't call transpose_densepose if you don't need
annos = [
self._transform_densepose(
utils.transform_instance_annotations(
obj,
transforms,
image_shape,
keypoint_hflip_indices=self.keypoint_hflip_indices),
transforms,
) for obj in dataset_dict.pop("annotations")
if obj.get("iscrowd", 0) == 0
]
if self.mask_on:
self._add_densepose_masks_as_segmentation(annos, image_shape)
instances = utils.annotations_to_instances(annos,
image_shape,
mask_format="bitmask")
densepose_annotations = [obj.get("densepose") for obj in annos]
if densepose_annotations and not all(v is None
for v in densepose_annotations):
instances.gt_densepose = DensePoseList(densepose_annotations,
instances.gt_boxes,
image_shape)
dataset_dict["instances"] = instances[instances.gt_boxes.nonempty()]
return dataset_dict
def __call__(self, dataset_dict):
"""
Args:
dataset_dict (dict): Metadata of one image, in Detectron2 Dataset format.
Returns:
dict: a format that builtin models in detectron2 accept
"""
dataset_dict = copy.deepcopy(dataset_dict) # it will be modified by code below
# USER: Write your own image loading if it's not from a file
# image = utils.read_image(dataset_dict["file_name"], format=self.img_format)
rota = 0
if self.rota_aug_on and dataset_dict["split"] != "val_mini" and dataset_dict["split"] != "test":
rotaed_aug = [0, 90, 180, 270]
rota = random.sample(rotaed_aug, 1)[0]
image = read_image(dataset_dict["file_name"], format=self.img_format, rota=rota)
utils.check_image_size(dataset_dict, image)
image, transforms = T.apply_transform_gens(self.tfm_gens, image)
image_shape = image.shape[:2] # h, w
# Pytorch's dataloader is efficient on torch.Tensor due to shared-memory,
# but not efficient on large generic data structures due to the use of pickle & mp.Queue.
# Therefore it's important to use torch.Tensor.
dataset_dict["image"] = torch.as_tensor(image.transpose(2, 0, 1).astype("float32"))
# Can use uint8 if it turns out to be slow some day
if not self.is_train:
dataset_dict.pop("annotations", None)
return dataset_dict
if "annotations" in dataset_dict:
# USER: Implement additional transformations if you have other types of data
annos = [
transform_dota_instance_annotations(
obj, image_shape, rota, transforms
)
for obj in dataset_dict.pop("annotations")
]
instances = dota_annotations_to_instances(
annos, image_shape
)
dataset_dict["instances"] = filter_empty_instances(instances)
return dataset_dict
def __call__(self, dataset_dict):
self.tfm_gens = []
dataset_dict = deepcopy(dataset_dict)
image = utils.read_image(dataset_dict["file_name"],
format=self.img_format)
utils.check_image_size(dataset_dict, image)
if self.is_train:
# Crop
'''print("Augmentation: ", "T.RandomCrop('relative', [0.8, 0.4])")
crop_gen = T.RandomCrop('relative', [0.8, 0.4])
self.tfm_gens.append(crop_gen)'''
# Horizontal flip
print("Augmentation: ",
"T.RandomFlip(prob=0.5, horizontal=True, vertical=False)")
flip_gen = T.RandomFlip(prob=0.5, horizontal=True, vertical=False)
self.tfm_gens.append(flip_gen)
image, transforms = T.apply_transform_gens(self.tfm_gens, image)
image_shape = image.shape[:2] # h, w
dataset_dict["image"] = torch.as_tensor(
np.ascontiguousarray(image.transpose(2, 0, 1)))
if not self.is_train:
dataset_dict.pop("annotations", None)
return dataset_dict
if "annotations" in dataset_dict:
for anno in dataset_dict["annotations"]:
if not self.mask_on:
anno.pop("segmentation", None)
annos = [
utils.transform_instance_annotations(
obj,
transforms,
image_shape,
keypoint_hflip_indices=self.keypoint_hflip_indices)
for obj in dataset_dict.pop("annotations")
if obj.get("iscrowd", 0) == 0
]
instances = utils.annotations_to_instances(
annos, image_shape, mask_format=self.mask_format)
if self.crop_gen and instances.has("gt_masks"):
instances.gt_boxes = instances.gt_masks.get_bounding_boxes()
dataset_dict["instances"] = utils.filter_empty_instances(instances)
return dataset_dict
def __call__(self, dataset_dict):
"""
Args:
dataset_dict (dict): Metadata of one image, in Detectron2 Dataset format.
Returns:
dict: a format that builtin models in detectron2 accept
"""
dataset_dict = copy.deepcopy(dataset_dict)
image = utils.read_image(dataset_dict['file_name'],
format=self.img_format)
utils.check_image_size(dataset_dict, image)
if self.crop_gen is None:
image, transforms = T.apply_transform_gens(self.tfm_gens, image)
elif np.random.rand() > 0.5:
image, transforms = T.apply_transform_gens(self.tfm_gens, image)
else:
image, transforms = T.apply_transform_gens(
self.tfm_gens[:-1] + self.crop_gen + self.tfm_gens[-1:], image)
image_shape = image.shape[:2]
dataset_dict['image'] = paddle.to_tensor(
np.ascontiguousarray(image.transpose(2, 0, 1)))
if not self.is_train:
dataset_dict.pop('annotations', None)
return dataset_dict
if 'annotations' in dataset_dict:
for anno in dataset_dict['annotations']:
if not self.mask_on:
anno.pop('segmentation', None)
anno.pop('keypoints', None)
annos = [
utils.transform_instance_annotations(obj, transforms,
image_shape)
for obj in dataset_dict.pop('annotations')
if obj.get('iscrowd', 0) == 0
]
instances = utils.annotations_to_instances(annos, image_shape)
dataset_dict['instances'] = utils.filter_empty_instances(instances)
return dataset_dict
def mapper(dataset_dict):
# Implement a mapper, similar to the default DatasetMapper, but with your own customizations
dataset_dict = copy.deepcopy(dataset_dict) # it will be modified by code below
image = utils.read_image(dataset_dict["file_name"], format="BGR")
image, transforms = T.apply_transform_gens([T.Resize((800, 800))], image)
dataset_dict["image"] = torch.as_tensor(image.transpose(2, 0, 1).astype("float32"))
annos = [
utils.transform_instance_annotations(obj, transforms, image.shape[:2])
for obj in dataset_dict.pop("annotations")
if obj.get("iscrowd", 0) == 0
]
instances = utils.annotations_to_instances(annos, image.shape[:2])
dataset_dict["instances"] = utils.filter_empty_instances(instances)
return dataset_dict
def mapper(dataset_dict):
dataset_dict = copy.deepcopy(dataset_dict)
image = utils.read_image(dataset_dict["file_name"], format="BGR")
image, transforms = T.apply_transform_gens([T.Resize((1152, 1152))], image)
dataset_dict["image"] = torch.as_tensor(
image.transpose(2, 0, 1).astype("float32"))
annos = [
transform_instance_annotations(obj, transforms, image.shape[:2])
for obj in dataset_dict.pop("annotations")
if obj.get("iscrowd", 0) == 0
]
instances = utils.annotations_to_instances_rotated(annos, image.shape[:2])
dataset_dict["instances"] = utils.filter_empty_instances(instances)
return dataset_dict
def __call__(self, dataset_dict):
ret = super().__call__(dataset_dict=dataset_dict)
numpy_image = dataset_dict["image"].permute(1, 2, 0).numpy()
for angle in self.angles:
rotate = RandomRotation(angle=angle, expand=True)
new_numpy_image, tfms = apply_transform_gens([rotate], np.copy(numpy_image))
torch_image = torch.from_numpy(np.ascontiguousarray(new_numpy_image.transpose(2, 0, 1)))
dic = copy.deepcopy(dataset_dict)
# In DatasetMapperTTA, there is a pre_tfm transform (resize or no-op) that is
# added at the beginning of each TransformList. That's '.transforms[0]'.
dic["transforms"] = TransformList(
[ret[-1]["transforms"].transforms[0]] + tfms.transforms
)
dic["image"] = torch_image
ret.append(dic)
return ret
def custom_mapper(dataset_dict, transform_list):
dataset_dict = copy.deepcopy(
dataset_dict) # it will be modified by code below
image = utils.read_image(dataset_dict["file_name"], format="BGR")
image, transforms = T.apply_transform_gens(transform_list, image)
dataset_dict["image"] = torch.as_tensor(
image.transpose(2, 0, 1).astype("float32"))
annos = [
utils.transform_instance_annotations(obj, transforms, image.shape[:2])
for obj in dataset_dict.pop("annotations")
]
instances = utils.annotations_to_instances(annos, image.shape[:2])
dataset_dict["instances"] = utils.filter_empty_instances(instances)
return dataset_dict
def __call__(self, dataset_dict):
"""
Args:
dataset_dict (dict): Metadata of one image, in Detectron2 Dataset format.
Returns:
dict: a format that builtin models in detectron2 accept
"""
dataset_dict = copy.deepcopy(
dataset_dict) # it will be modified by code below
image = utils.read_image(dataset_dict["file_name"],
format=self.img_format)
utils.check_image_size(dataset_dict, image)
image, transforms = T.apply_transform_gens(self.augmentation, image)
image_shape = image.shape[:2] # h, w
dataset_dict["image"] = torch.as_tensor(
image.transpose(2, 0, 1).astype("float32"))
if not self.is_train:
dataset_dict.pop("annotations", None)
return dataset_dict
annos = [
utils.transform_instance_annotations(obj, transforms,
image.shape[:2])
for obj in dataset_dict.pop("annotations")
]
dataset_dict["instances"] = utils.annotations_to_instances(
annos, image.shape[:2])
# # USER: Implement additional transformations if you have other types of data
# # USER: Don't call transpose_densepose if you don't need
# annos = [
# self._transform_densepose(
# utils.transform_instance_annotations(
# obj, transforms, image_shape, keypoint_hflip_indices=self.keypoint_hflip_indices
# ),
# transforms,
# )
# for obj in dataset_dict.pop("annotations")
# if obj.get("iscrowd", 0) == 0
# ]
# instances = utils.annotations_to_instances(annos, image_shape, mask_format="bitmask")
# dataset_dict["instances"] = instances[instances.gt_boxes.nonempty()]
return dataset_dict
def __call__(self, dataset_dict):
"""
Args:
dataset_dict (dict): Metadata of ONE video, in Detectron2 Dataset format.
Returns:
dict: a format that builtin models in detectron2 accept
"""
dataset_dict = copy.deepcopy(dataset_dict) # it will be modified by code below
if self.is_train:
dataset_dict = # TODO: sample a fixed number of frames
new_dataset_dict = []
for item in dataset_dict:
image = utils.read_image(item["filename"], format=self.img_format)
utils.check_image_size(item, image)
# TODO: SSD random crop
image, transforms = T.apply_transform_gens(
self.tfm_gens[:-1] + self.crop_gen + self.tfm_gens[-1:], image
)
image_shape = image.shape[:2] # h, w
# Pytorch's dataloader is efficient on torch.Tensor due to shared-memory,
# but not efficient on large generic data structures due to the use of pickle & mp.Queue.
# Therefore it's important to use torch.Tensor.
image = torch.as_tensor(np.ascontiguousarray(image.transpose(2, 0, 1)))
sample = {"image": image}
if not self.is_train:
new_dataset_dict.append(sample)
continue
# USER: Implement additional transformations if you have other types of data
boxes = [
utils.transform_instance_annotations(box, transforms, image_shape)
for box in item["boxes"]
]
instances = #
sample["instances"] = instances
new_dataset_dict.append(sample)
return new_dataset_dict
def test_apply_rotated_boxes(self):
np.random.seed(125)
cfg = get_cfg()
is_train = True
transform_gen = detection_utils.build_transform_gen(cfg, is_train)
image = np.random.rand(200, 300)
image, transforms = T.apply_transform_gens(transform_gen, image)
image_shape = image.shape[:2] # h, w
assert image_shape == (800, 1200)
annotation = {"bbox": [179, 97, 62, 40, -56]}
boxes = np.array([annotation["bbox"]], dtype=np.float64) # boxes.shape = (1, 5)
transformed_bbox = transforms.apply_rotated_box(boxes)[0]
expected_bbox = np.array([484, 388, 248, 160, 56], dtype=np.float64)
err_msg = "transformed_bbox = {}, expected {}".format(transformed_bbox, expected_bbox)
assert np.allclose(transformed_bbox, expected_bbox), err_msg
def __call__(self, dataset_dict):
"""
Args:
dataset_dict (dict): Metadata of one image, in Detectron2 Dataset format.
Returns:
dict: a format that builtin models in detectron2 accept
"""
dataset_dict = copy.deepcopy(
dataset_dict) # it will be modified by code below
image = utils.read_image(dataset_dict["file_name"],
format=self.img_format)
utils.check_image_size(dataset_dict, image)
assert "sem_seg_file_name" in dataset_dict
image, transforms = T.apply_transform_gens(self.tfm_gens, image)
if self.is_train:
with PathManager.open(dataset_dict.pop("sem_seg_file_name"),
"rb") as f:
sem_seg_gt = Image.open(f)
sem_seg_gt = np.asarray(sem_seg_gt, dtype="uint8")
sem_seg_gt = transforms.apply_segmentation(sem_seg_gt)
if self.crop_gen:
image, sem_seg_gt = crop_transform(
image,
sem_seg_gt,
self.crop_gen,
self.single_category_max_area,
self.ignore_value,
)
dataset_dict["sem_seg"] = torch.as_tensor(
sem_seg_gt.astype("long"))
# Pytorch's dataloader is efficient on torch.Tensor due to shared-memory,
# but not efficient on large generic data structures due to the use of pickle & mp.Queue.
# Therefore it's important to use torch.Tensor.
dataset_dict["image"] = torch.as_tensor(
np.ascontiguousarray(image.transpose(2, 0, 1)))
if not self.is_train:
dataset_dict.pop("sem_seg_file_name", None)
return dataset_dict
return dataset_dict
def __call__(self, dataset_dict):
"""
Args:
dict: a dict in standard model input format. See tutorials for details.
Returns:
list[dict]:
a list of dicts, which contain augmented version of the input image.
The total number of dicts is ``len(min_sizes) * (2 if flip else 1)``.
Each dict has field "transforms" which is a TransformList,
containing the transforms that are used to generate this image.
"""
numpy_image = dataset_dict["image"].permute(1, 2, 0).numpy()
shape = numpy_image.shape
orig_shape = (dataset_dict["height"], dataset_dict["width"])
if shape[:2] != orig_shape:
# It transforms the "original" image in the dataset to the input image
pre_tfm = ResizeTransform(orig_shape[0], orig_shape[1], shape[0],
shape[1])
else:
pre_tfm = NoOpTransform()
# Create all combinations of augmentations to use
tfm_gen_candidates = [] # each element is a list[TransformGen]
for min_size in self.min_sizes:
resize = ResizeShortestEdge(min_size, self.max_size)
tfm_gen_candidates.append([resize]) # resize only
if self.flip:
flip = RandomFlip(prob=1.0)
tfm_gen_candidates.append([resize, flip]) # resize + flip
# Apply all the augmentations
ret = []
for tfm_gen in tfm_gen_candidates:
new_image, tfms = apply_transform_gens(tfm_gen,
np.copy(numpy_image))
torch_image = torch.from_numpy(
np.ascontiguousarray(new_image.transpose(2, 0, 1)))
dic = copy.deepcopy(dataset_dict)
dic["transforms"] = pre_tfm + tfms
dic["image"] = torch_image
ret.append(dic)
return ret
def customMapper(dataset_dict):
dataset_dict = copy.deepcopy(dataset_dict)
image = utils.read_image(dataset_dict["file_name"], format="BGR")
transform_list = [
T.Resize((600, 800)),
T.RandomFlip(prob=0.6, horizontal=True, vertical=False),
T.RandomFlip(prob=0.6, horizontal=False, vertical=True),
]
image, transforms = T.apply_transform_gens(transform_list, image)
dataset_dict["image"] = torch.as_tensor(image.transpose(2, 0, 1).astype("float32"))
annos = [
utils.transform_instance_annotations(obj, transforms, image.shape[:2])
for obj in dataset_dict.pop("annotations")
if obj.get("iscrowd", 0) == 0
]
instances = utils.annotations_to_instances(annos, image.shape[:2])
dataset_dict["instances"] = utils.filter_empty_instances(instances)
return dataset_dict
def __call__(self, dataset_dict):
"""
Args:
dataset_dict (dict): Metadata of one image, in Detectron2 Dataset format.
Returns:
dict: a format that builtin models in detectron2 accept
"""
# Implement a mapper, similar to the default DatasetMapper, but with own customizations
dataset_dict = copy.deepcopy(
dataset_dict) # it will be modified by code below
image = utils.read_image(dataset_dict["file_name"], format="BGR")
# Custom augs to be used while training
# Only HFlip and Resize are supported for rotated_boxes
augs = [T.RandomFlip(0.4, horizontal=True,
vertical=False)] #[T.RandomRotation([0,90])]
if self.is_train:
tfm_gens = self.tfm_gens + augs
else:
tfm_gens = self.tfm_gens
logging.getLogger(__name__).info("Original Augmentation: " +
str(self.tfm_gens))
logging.getLogger(__name__).info("Updated Augmentation List: " +
str(tfm_gens))
image, transforms = T.apply_transform_gens(tfm_gens, image)
dataset_dict["image"] = torch.as_tensor(
image.transpose(2, 0, 1).astype("float32"))
for a in dataset_dict['annotations']:
a['bbox'] = transforms.apply_rotated_box(np.asarray(
[a['bbox']]))[0].tolist()
annos = dataset_dict['annotations']
instances = utils.annotations_to_instances_rotated(
annos, image.shape[:2])
dataset_dict["instances"] = utils.filter_empty_instances(instances)
return dataset_dict
