Load a config file and merge it into the default options.

Visualize segmentations nicely. Based on code from: https://github.com/roytseng-tw/Detectron.pytorch/blob/master/lib/utils/vis.py @p

Return an activation function given a string

Compute Overlap and Boundary Precision, Recall, F-measure Also compute #objects detected, #confident objects detected, #GT objects.

Add config for MEAN SHIFT MASK_FORMER.

Build a pixel decoder from `cfg.MODEL.MASK_FORMER.PIXEL_DECODER_NAME`.

Given an image array, add the hue, saturation and luminosity to the image

Combine several bit masks [N, H, W] into a mask [H,W], e.g. 8*480*640 tensor becomes a numpy array of 480*640. [[1,0,0], [0,1,0]] = > [2,

Combine several bit masks [N, H, W] into a mask [H,W], e.g. 8*480*640 tensor becomes a numpy array of 480*640. [[1,0,0], [0,1,0]] = > [2,

Extract objects with high prediction scores.

2x2 max pooling

Normalizes the descriptor into RGB color space :param res: numpy.array [H,W,D] Output of the network, per-pixel dense descriptor

From a segmentation, compute a binary boundary map with 1 pixel wide boundaries. The boundary pixels are offset by 1/2 pixel towards the

NOTE: this interface is experimental. Args: input_shape: shapes (channels and stride) of the input features c

NOTE: this interface is experimental. Args: input_shape: shapes (channels and stride) of the input features t

Return an activation function given a string

Return an activation function given a string

Resizes the input image to allow proper feature alignment during the forward propagation. Resizes the input image to a closest multiple of `o

Build a instance embedding branch from `cfg.MODEL.INS_EMBED_HEAD.NAME`.

Compute the indexes for the lowest-cost pairings between rows and columns in the database. Returns a list of (row, column) tuples

3x3 convolution with padding

Selects seeds that are as far away as possible @param X: a [n x d] torch.FloatTensor of d-dim unit vectors @param num_seeds: number

Clear all covered matrix cells

Args: shape: (h, w) tuple or a int interp: PIL interpolation method

Create an *n*x*n* matrix, populating it with the specific value.

Get an imdb (image database) by name.

Return the directory where experimental artifacts are placed. A canonical path is built using the name from an imdb and a network (if not Non

r""" Modified from multi_head_attention_forward in PyTorch.nn.functional Args: query, key, value: map a query and a set of key-value

inpaint the input depth where the value is equal to zero Args: depth ([np.uint8]): normalized depth array [H, W, 3] (0 ~ 255) fa

Runs mean shift with carefully selected seeds @param X: a [n x d] torch.FloatTensor of d-dim unit vectors @param dist_threshold: par

normalize the input depth (mm) and return depth image (0 ~ 255) Args: depth ([np.float]): depth array [H, W] (mm) min_val (float,

Process foreground_labels - Map the foreground_labels to {0, 1, ..., K-1} @param foreground_labels: a [H x W] numpy arra

Process foreground_labels - Map the foreground_labels to {0, 1, ..., K-1} @param foreground_labels: a [H x W] numpy arra

Process foreground_labels - Map the foreground_labels to {0, 1, ..., K-1} @param foreground_labels: a [H x W] numpy arra

Process foreground_labels - Map the foreground_labels to {0, 1, ..., K-1} @param foreground_labels: a [H x W] numpy arra

Process labels - Map the labels to [H x W x num_instances] numpy array

Process labels - Map the labels to [H x W x num_instances] numpy array

Create configs and perform basic setups.

Convert the model into training mode while keep layers freezed.

Args: x: (B, H, W, C) window_size (int): window size Returns: windows: (num_windows*B, window_size, window_size, C)

Erase all prime markings

Find the first uncovered element with value 0

Find the first prime element in the specified row. Returns the column index, or -1 if no starred element was found.