Load a yaml config file and merge it into the global config.

Reduce the loss dictionary from all processes so that process with rank 0 has the averaged results. Returns a dict with the same fields as

3x3 convolution with padding

Input: x [.., c, M_x] y [.., c, M_y] Return: ret [.., M_x, M_y]

Merge config keys, values in a list (e.g., from command line) into the global config. For example, `cfg_list = ['TEST.NMS', 0.5]`.

Checks that `value_a`, which is intended to replace `value_b` is of the right type. The type is correct if it matches exactly or is one of a few

Decodes a raw config value (e.g., from a yaml config files or command line argument) into a Python object.

Constructs a linear kernel matrix between A and B. We assume that each row in A and B represents a d-dimensional feature vector.

:param size: size of the height/width :param ratio: generate grid size/ even divided grid size :param num_grid: number of gri

Helper function to synchronize (barrier) among all processes when using distributed training

Merge config dictionary a into config dictionary b, clobbering the options in b whenever they are also specified in a.

Computes an inverse of each matrix in the batch. Pytorch 0.4.1 does not support batched matrix inverse. Hence, we are solving

Constructs a linear kernel matrix between A and B. We assume that each row in A and B represents a d-dimensional feature vector.

query B x D x H x W support Nc x Nk x Ds x Hs x Ws (#CLASSES x #SHOT x #DIMENSIONS)