Extract metadata for image i, For efficient data loading.

Load input image and preprocess for Caffe: - cast to float - switch channels RGB -> BGR - subtract mean - transpose to channel x

Expands an array of integers in one-hot encoding by adding a new last dimension, leaving zeros everywhere except for the nth dimensio

Normalize answers from a given answer json in the usual VQA format.

Tokenize and normalize questions from a given question json in the usual VQA format.

Returns an array of primary category numbers for each label, where the first category listed in ds.category_names is given category n

Returns the TopN to an empty state.

Constructs a ResNet-152 model. Args: pretrained (bool): If True, returns a model pre-trained on ImageNet

Resolves a full segmentation, potentially in a differenct process, for efficient multiprocess data loading.

Zooms a 2d or 3d segmentation to the given dims, using nearest neighbor.

The number of images in this data set.

Returns a single batch as an array of tensors, one per category.

Repeat the same feature vector over all spatial positions of a given feature map. The feature vector should have the same batch size and numb

Compares Captions by score.

Initializes the Caption. Args: sentence: List of word ids in the caption. state: Model state after generating the previou

Initializes the generator. Args: model: recurrent model, with inputs: (input, state) and outputs len(vocab) values beam_s

Constructor arguments: segmentation: The AbstractSegmentation to load. split: None for no filtering, or 'train' or 'val' etc.

Create a mapping from a COCO image id into the corresponding index into the h5 file

All English synonyms for the given label

Returns the number of images touched by a category.

Crops image_data to the specified receptive field. Together fieldmap and location specify a receptive field on the image, which may overlap t