vqa_benchmarking_backend.datasets.VQADataset

Module Contents

Classes

VQADataSample

Class describing one data sample of the VQA2 dataset

VQADataset

Class describing the VQA2 dataset

Functions

preprocess_question(question: str) → List[str]

Removes punctuation and make everything lower case

load_img(path: str, transform=None) → numpy.ndarray

Loads an image using module cv2

load_img_feats(path: str) → torch.FloatTensor

Loads a numpy array containing image features

preprocess_answer(answer: str) → str

Removes punctuation

answer_score(num_humans) → float

Calculates VQA score in [0,1] depending on number of humans having given the same answer
vqa_benchmarking_backend.datasets.VQADataset.preprocess_question(question: str) → List[str]

Removes punctuation and make everything lower case

vqa_benchmarking_backend.datasets.VQADataset.load_img(path: str, transform=None) → numpy.ndarray

Loads an image using module cv2

vqa_benchmarking_backend.datasets.VQADataset.load_img_feats(path: str) → torch.FloatTensor

Loads a numpy array containing image features

vqa_benchmarking_backend.datasets.VQADataset.preprocess_answer(answer: str) → str

Removes punctuation

vqa_benchmarking_backend.datasets.VQADataset.answer_score(num_humans) → float

Calculates VQA score in [0,1] depending on number of humans having given the same answer

class vqa_benchmarking_backend.datasets.VQADataset.VQADataSample(question_id: str, question: str, answers: Dict[str, float], image_id: str, image_path: str, image_feat_path: str, image_transform=None)

Bases: vqa_benchmarking_backend.datasets.dataset.DataSample

Class describing one data sample of the VQA2 dataset Inheriting from DataSample

property image(self) → numpy.ndarray

Returns the image, if not present it loads it from self._image_path

property question_tokenized(self) → List[str]

Returns tokenized question

property question(self) → str

Returns full question

class vqa_benchmarking_backend.datasets.VQADataset.VQADataset(val_question_file: str, val_annotation_file: str, answer_file: str, img_dir, img_feat_dir, name: str, transform=None, load_img_features=False, dataset_fraction: float = 0.05, random_seed: int = 12345)

Bases: vqa_benchmarking_backend.datasets.dataset.DiagnosticDataset

Class describing the VQA2 dataset Inheriting from DiagnosticDataset

__len__(self)
_load_data(self, question_file: str, annotation_file: str, dataset_fraction: float, random_seed: int) → Tuple[List[vqa_benchmarking_backend.datasets.dataset.DataSample], Dict[str, vqa_benchmarking_backend.datasets.dataset.DataSample], vqa_benchmarking_backend.utils.vocab.Vocabulary, vqa_benchmarking_backend.utils.vocab.Vocabulary]

Loads data from VQA json files Returns:

  • data: list of VQADataSample

  • qid_to_sample: mapping of question id to data sample

  • question_vocab: Vocabulary of all unique words occuring in the data

  • answer_vocab: Vocabulary of all unique answers

__getitem__(self, index) → vqa_benchmarking_backend.datasets.dataset.DataSample

Returns a data sample

get_name(self) → str

Returns the name of the dataset, required for file caching

class_idx_to_answer(self, class_idx: int) → Union[str, None]

Get the answer string for a given class index from the self.idx2ans dictionary