Visual Question Answering

Visual Question Answering is the task of answering open-ended questions based on an image. They output natural language responses to natural language questions.


What is in this image?

Visual Question Answering Model

About Visual Question Answering

Use Cases

Aid the Visually Impaired Persons

VQA models can be used to reduce visual barriers for visually impaired individuals by allowing them to get information about images from the web and the real world.


VQA models can be used to improve experiences at museums by allowing observers to directly ask questions they interested in.

Improved Image Retrieval

Visual question answering models can be used to retrieve images with specific characteristics. For example, the user can ask "Is there a dog?" to find all images with dogs from a set of images.

Video Search

Specific snippets/timestamps of a video can be retrieved based on search queries. For example, the user can ask "At which part of the video does the guitar appear?" and get a specific timestamp range from the whole video.

Task Variants

Video Question Answering

Video Question Answering aims to answer questions asked about the content of a video.


You can infer with Visual Question Answering models using the vqa (or visual-question-answering) pipeline. This pipeline requires the Python Image Library (PIL) to process images. You can install it with (pip install pillow).

from PIL import Image
from transformers import pipeline

vqa_pipeline = pipeline("visual-question-answering")

image =  Image.open("elephant.jpeg")
question = "Is there an elephant?"

vqa_pipeline(image, question, top_k=1)
#[{'score': 0.9998154044151306, 'answer': 'yes'}]

Useful Resources

The contents of this page are contributed by Bharat Raghunathan and Jose Londono Botero.

Visual Question Answering demo
Visual Question Answering
Drag image file here or click to browse from your device
This model can be loaded on the Inference API on-demand.
Models for Visual Question Answering
Browse Models (6)

Note Robust Visual Question Answering model trained on the VQAv2 dataset.

Datasets for Visual Question Answering

Note A widely used dataset containing questions (with answers) about images.

Note A dataset to benchmark visual reasoning based on text in images.

Metrics for Visual Question Answering
Accuracy is the proportion of correct predictions among the total number of cases processed. It can be computed with: Accuracy = (TP + TN) / (TP + TN + FP + FN) Where: TP: True positive TN: True negative FP: False positive FN: False negative
wu-palmer similarity
Measures how much a predicted answer differs from the ground truth based on the difference in their semantic meaning.