Index instance tokens for uniquely identifying instances across samples

Args: x (dict): one dataset dict with "image" field being a CHW tensor Returns: dict: one output dict

Args: x (dict): one dataset dict with "image" field being a CHW tensor Returns: dict: one output dict

Build calibration table from dataset. Calibration details here: https://github.com/NVIDIA/DIGITS/blob/v4.0.0-rc.3/digits/extensions/da

Adapted from AdelaiDet: https://github.com/aim-uofa/AdelaiDet/blob/master/adet/utils/comm.py

Parameters ---------- quat: Tensor (N, 4). Batch of (allocentric) quaternions. proj_ctr: Tensor (N, 2). Projected ce

Get the smallest angle difference between 2 angles: the angle from y to x. Parameters ---------- x: float To angle. y: float

Create an :class:`Instances` object used by the models, from instance annotations in the dataset dict. Args: annos (list[dict]):

Compute overlap in BEV

Compute 3D box IoU

Parameters ---------- boxes3d: Boxes3D 3D boxes in camera frame. pose_cam_global: Pose Transformation from sensor (c

Changes from the original function: - Move `RandomCrop` augmentation here; it's originally in dataset_mapper https://github.

Build a batched dataloader for training. Args: dataset (torch.utils.data.Dataset): map-style PyTorch dataset. Can be indexed.

Copied from detectron2.utils.visualizer.py ------------------------------------------- Depending on the brightness_factor, gives a light

This method compares predictions to the ground truth in terms of scale. It is equivalent to intersection over union (IOU) between the two box

Compute TP/FP statistics. Modified from https://github.com/sshaoehuai/PointRCNN/blob/master/tools/kitti_object_eval_python/eval.py

Compute TP/FP statistics. Modified from https://github.com/sshaoehuai/PointRCNN/blob/master/tools/kitti_object_eval_python/eval.py

Convert an instance detection/segmentation or keypoint detection dataset in detectron2's standard format into COCO json format. Generic

Copy of https://github.com/facebookresearch/detectron2/blob/master/detectron2/data/datasets/coco.py#L445, but removing the `file_lock`.

Wrapper of D2's 'Visualizer.draw_instance_predictions()' to render GT instances.

Copied from Visualizer.draw_text() ----------------------------------- Args: text (str): class label position (tuple): a

3x3 convolution with padding

Color interior of polygon with alpha-blending. This function modified input in place.

Initialize pose from rotation matrix and translation vector. Parameters ---------- rotation_matrix : np.ndarray 3

Compute TP/FP statistics. Taken from https://github.com/sshaoehuai/PointRCNN/blob/master/tools/kitti_object_eval_python/eval.py without change

Gather python dictionaries from all workers to the rank=0 worker. Assumption: the keys of `dikt` are disjoint across all workers. If rank =

Get physical pixel size of pinhole cameras. ((1 / fx) ** 2 + (1 / fy) ** 2)).sqrt() Parameters ---------- cams: PerspectiveCameras

Adapted from d2.utils.env:seed_all_rng()

Given an array of precision, recall values, sample evenly along the recall range, and interpolate the precision based on AP from section 6 fro

Get thresholds from a set of scores, up to num sample points Parameters ---------- score: np.ndarray Numpy array of scores for pr

Parameters ---------- xy: Tensor xy coordinates. shape=(N, ..., 2) E.g., (N, 2) or (N, K, 2) or (N, H, W, 2) Returns

Return the sum of a list of objects. can be int, float, torch.Tensor, np.ndarray, etc can be used for adding losses

If the checkpoint is an S3 or https path, the main process download the weight under, by default, `/tmp/`. NOTE: All workers must update `cfg.MOD

Wrapper function for cleaning data before computing metrics.

Render the bounding box on the image. NOTE: CV2 renders in place. Parameters ---------- box3d: GenericBoxes3D image: np.uint8 array

Render bounding box 3d in BEV perspective. Parameters ---------- bboxes3d: GenericBoxes3D 3D annotations in the s

Visualize the frustrum of camera by drawing two lines connecting the camera center and top-left / top-right corners of image plane.

Render point cloud in BEV perspective. Parameters ---------- point_cloud: numpy array with shape (N, 3) 3D cloud

Reset the canvas to a blank image with guideline circles of various radii.

Adapted from detectron2: https://github.com/facebookresearch/detectron2/blob/master/detectron2/engine/launch.py#L85

Return the first item of the given list if the list only contains one item. usually used in args parsing

Adapted from: https://github.com/facebookresearch/detectron2/blob/master/detectron2/data/detection_utils.py#L254 The changes from origina

Verify that the format/dimensions for the annotations are correct. Keys correspond to defintions here: https://github.com/NVIDIA/DIGIT

Args: dict: a dict in standard model input format. See tutorials for details. Returns: list[dict]:

Same input/output format as :meth:`DD3D`

Same input/output format as :meth:`NuscenesDD3D`

Add 'pred_attribute', 'pred_speed' to Instances in 'pred_instances'.