Tumblr Animated GIFs With Crowdsourced Descriptions

The Tumblr GIF (TGIF) dataset contains 100,000 animated GIFs and 120,000 descriptive sentences. GIFs were collected from randomly selected Tumblr posts published between May and June 2015, with sentences gathered via a crowdsourced annotation interface. It is designed for evaluating animated GIF and video description techniques.

Use Cases

• Train a model to generate descriptive sentences for animated GIFs using the 120,000 provided captions.
• Evaluate video description models on the test split, which contains three sentences per GIF for assessment.
• Analyze the relationship between visual content in GIFs and the language used in their crowdsourced descriptions.

Strengths

• Contains 100,000 animated GIFs, providing a substantial collection for visual-language tasks.
• Includes 120,000 descriptive sentences, offering multiple captions for many GIFs.
• Sentences were collected via a carefully designed crowdsourcing interface intended to ensure high quality.

Limitations

• GIFs are sourced only from Tumblr posts over a two-month period in 2015, which may limit temporal and platform diversity.
• The dataset's primary use is for description evaluation, which may limit its applicability for other generative or analytical tasks.
• Data is provided as URLs to GIFs, which may lead to broken links over time.

Provenance

Source

Tumblr posts.

Collection Method

GIFs collected from randomly selected Tumblr posts; sentences collected via crowdsourcing with a designed annotation interface.

Time Range

May to June 2015.

Freshness

Last updated on the platform in October 2022, but the underlying data is from 2015.

Geography

null