Spaces:
Running
on
Zero
Running
on
Zero
File size: 4,126 Bytes
61c2d32 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 |
# -*- coding: utf-8 -*-
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
import numpy as np
import torch
# fmt: off
from detectron2.data.detection_utils import \
annotations_to_instances as base_annotations_to_instances
from detectron2.data.detection_utils import \
transform_instance_annotations as base_transform_instance_annotations
# fmt: on
def annotations_to_instances(annos, image_size, sample_points=0):
"""
Create an :class:`Instances` object used by the models,
from instance annotations in the dataset dict.
Args:
annos (list[dict]): a list of instance annotations in one image, each
element for one instance.
image_size (tuple): height, width
sample_points (int): subsample points at each iteration
Returns:
Instances:
It will contain fields "gt_boxes", "gt_classes",
"gt_point_coords", "gt_point_labels", if they can be obtained from `annos`.
This is the format that builtin models with point supervision expect.
"""
target = base_annotations_to_instances(annos, image_size)
assert ("point_coords" in annos[0]) == ("point_labels" in annos[0])
if len(annos) and "point_labels" in annos[0]:
point_coords = []
point_labels = []
for i, _ in enumerate(annos):
# Already in the image coordinate system
point_coords_wrt_image = np.array(annos[i]["point_coords"])
point_labels_wrt_image = np.array(annos[i]["point_labels"])
if sample_points > 0:
random_indices = np.random.choice(
point_coords_wrt_image.shape[0],
sample_points,
replace=point_coords_wrt_image.shape[0] < sample_points,
).astype(int)
point_coords_wrt_image = point_coords_wrt_image[random_indices]
point_labels_wrt_image = point_labels_wrt_image[random_indices]
assert point_coords_wrt_image.shape[0] == point_labels_wrt_image.size
point_coords.append(point_coords_wrt_image)
point_labels.append(point_labels_wrt_image)
point_coords = torch.stack([torch.from_numpy(x) for x in point_coords])
point_labels = torch.stack([torch.from_numpy(x) for x in point_labels])
target.gt_point_coords = point_coords
target.gt_point_labels = point_labels
return target
def transform_instance_annotations(
annotation, transforms, image_size, *, keypoint_hflip_indices=None
):
"""
Apply transforms to box, and point annotations of a single instance.
It will use `transforms.apply_box` for the box, and
`transforms.apply_coords` for points.
Args:
annotation (dict): dict of instance annotations for a single instance.
It will be modified in-place.
transforms (TransformList or list[Transform]):
image_size (tuple): the height, width of the transformed image
keypoint_hflip_indices (ndarray[int]): see `create_keypoint_hflip_indices`.
Returns:
dict:
the same input dict with fields "bbox", "point_coords", "point_labels"
transformed according to `transforms`.
The "bbox_mode" field will be set to XYXY_ABS.
"""
annotation = base_transform_instance_annotations(
annotation, transforms, image_size, keypoint_hflip_indices
)
assert ("point_coords" in annotation) == ("point_labels" in annotation)
if "point_coords" in annotation and "point_labels" in annotation:
point_coords = annotation["point_coords"]
point_labels = np.array(annotation["point_labels"]).astype(float)
point_coords = transforms.apply_coords(point_coords)
# Set all out-of-boundary points to "unlabeled"
inside = (point_coords >= np.array([0, 0])) & (point_coords <= np.array(image_size[::-1]))
inside = inside.all(axis=1)
point_labels[~inside] = -1
annotation["point_coords"] = point_coords
annotation["point_labels"] = point_labels
return annotation
|