Find previous filtering nodes that might have text associated. The full filter may contain UNION nodes, which means that there might be severa

returns the iou matrix between two sets of boxes The boxes are expected in the format [top_left_x, top_left_y, w, h]

Accepts a list of bounding boxes in coco format (xmin,ymin, width, height) Returns the list of boxes in pascal format (xmin,ymin,xmax,ymax) T

Helper function to build the list of evaluators for a given dataset

Helper function to build the list of evaluators for a given dataset

Run per image evaluation on given images and store results (a list of dict) in self.eval_imgs.

Given a set of boxes, returns a dict containing clusters of boxes that are highly overlapping. For optimization, return None if none of the boxes

This functions computes the spans after reduction of the caption to it's normalized version For example, if the caption is "There is a man wearing

This functions accepts a dictionary that contains all the datapoints from a given id. These datapoints are assumed to come from the same image sub

Get a balanced subset containing the target fraction of the full dataset

Returns: The number of processes in the process group

Equivalent to nn.functional.interpolate, but with support for empty channel sizes.

Loads the CLEVR templates from disk

r""" Args: query, key, value: map a query and a set of key-value pairs to an output. See "Attention Is All You Need" for more

This function normalizes whitespaces, removing duplicates and converting all to standard spaces

Extract the bounding box of a given object as a list

Parse the program of the given question and return the objects, along with the list of tokens they correspond to from the scene necessary to compute t

Retrieve bounding boxes associated to given objs in given scene. This assumes non-REFCLEVR annotations (and thus can't provide a mask)

Retrieve bounding boxes associated to given objs in given scene. This assumes non-REFCLEVR annotations (and thus can't provide a mask)

Retrieve bounding boxes and segmentation masks associated to given objs in given scene This assumes REFCLEVR annotations

Retrieve bounding boxes and segmentation masks associated to given objs in given scene This assumes REFCLEVR annotations

Wrapper function which calculates the results.

Returns True if the given spans intersect with any in the given list

Returns True if the given spans intersect with any in the given list

Converts a set of boxes in [top_left_x, top_left_y, bottom_right_x, bottom_right_y] format to [top_left_x, top_left_y, width, height] format

Params for LVIS evaluation API.

Return a process group based on gloo backend, containing all the ranks The result is cached.

Return the boxes corresponding to the smallest enclosing box containing all the provided boxes The boxes are expected in [x1, y1, x2, y2] for

Adjust the lr according to the schedule. Args: Optimizer: torch optimizer to update. epoch(int): number of the current epoch.

Return intersection-over-union (Jaccard index) of boxes. Both sets of boxes are expected to be in ``(x1, y1, x2, y2)`` format with ``0 <

This build a regex that will match questions from the given text template. Additionally, tokens that are input of the node_id will be put in a

This build a regex that will match refexps from the given text template. Additionally, tokens that are input of the node_id will be put in a r

Transform a CLEVR text template into a regex that matches it

Transform a CLEVR text template into a regex that matches it

Yield successive n-sized chunks from lst.