Same as add just more pythonic.

Copies this config into a new config object, making the changes given by new_config_dict.

Returns the average of the elements in the window.

Adds an element to the window, removing the earliest element if necessary.

Copies new_config_dict into this config object. Note: new_config_dict can also be a config object.

Resets the MovingAverage to its initial state.

Here one layer means a string of n blocks.

Sanitizes the input coordinates so that x1 < x2, x1 != x2, x1 >= 0, and x2 <= image_size. Also converts from relative to absolute coordinates

Here one layer means a string of n Bottleneck blocks.

Compute the jaccard overlap of two sets of boxes. The jaccard overlap is simply the intersection over union of two boxes. Here we operate on

Postprocesses the output of Yolact on testing mode into a format that makes sense, accounting for all the possible configuration settings.

Note: If undo_transform=False then im_h and im_w are allowed to be None.

Sets the active config. Works even if cfg is already imported!

Implements a conv, activation, then batch norm. Arguments are passed into the conv layer.

Adapted from torchvision.models.mobilenet._make_divisible.

Convert prior_boxes to (cx, cy, w, h) representation for comparison to center-size form ground truth data. Args: boxes: (tensor) poin

Decode predicted bbox coordinates using the same scheme employed by Yolov2: https://arxiv.org/pdf/1612.08242.pdf b_x = (sigmoid(pred

Warp an image or feature map with optical flow Args: x (Tensor): size (n, c, h, w) flow (Tensor): size (n, 2, h, w), values range

Note: config should be config.name.

Loads weights from a compressed save file.

A helper function to take a config setting and turn it into a network. Used by protonet and extrahead. Returns (network, out_channels)

Saves the model's weights using compression because the file sizes were getting too big.

Sets the dataset of the current config.

A layer is a combination of inverted residual blocks

Note that bbox should be a list or tuple of (x1, y1, x2, y2)

The segmentation should be the full mask, the size of the image and with size [h, w].

Constructs a backbone given a backbone config object (see config.py).

"Crop" predicted masks by zeroing out everything not in the predicted bbox. Vectorized by Chong (thanks Chong). Args: - masks sh

Perform nms for only the max scoring class that isn't background (class 0)

Adapted from https://discuss.pytorch.org/t/how-to-train-with-frozen-batchnorm/12106/8

transformed_cat_id is [0,80) as indices in cfg.dataset.class_names

Initialize weights for training.

Note that priors are [x,y,width,height] where (x,y) is the center of the box.

Inputs inputs are matricies of size _ x N. Output is size _1 x _2. Note: if iscrowd is True, then mask2 should be the crowd.

Convert prior_boxes to (xmin, ymin, xmax, ymax) representation for comparison to point form ground truth data. Args: boxes: (tensor)

Prepare inverted table for category id lookup given a coco cats object.

Returns a list of APs for this image, with each element being for a class

Args: index (int): Index Returns: tuple: Tuple (image, target, masks, height, width, crowd).

Args: index (int): Index Returns: tuple: Tuple (image, target, masks, height, width, crowd).

Start timing the specific function. Note: If use_stack is True, only one timer can be active at a time. Once you stop this timer, t

If use_stack is True, this will stop the currently running timer and restore the previous timer on the stack if that exists. Note if use_stac

Takes a transformed image tensor and returns a numpy ndarray that is untransformed. Arguments w and h are the original height and width of th

Each layer is a sequence of conv layers usually preceded by a max pooling. Adapted from torchvision.models.vgg.make_layers.

Call this once per image.

Add a downsample layer to the backbone as per what SSD does.

Just a quick and dirty hash function for doing a deterministic shuffle based on image_id. Source: https://stackoverflow.com/questions/66