Dataset Card for ICQ-Highlight

Dataset Details

A dataset for Localizing Events in Videos with Multimodal Queries (Reference image + refinement text)

Dataset Description

Video understanding is a pivotal task in the digital era, yet the dynamic and multievent nature of videos makes them labor-intensive and computationally demanding to process. Thus, localizing a specific event given a semantic query has gained importance in both user-oriented applications like video search and academic research into video foundation models. A significant limitation in current research is that semantic queries are typically in natural language that depicts the semantics of the target event. This setting overlooks the potential for multimodal semantic queries composed of images and texts. To address this gap, we introduce a new benchmark, ICQ, for localizing events in videos with multimodal queries, along with a new evaluation dataset ICQ-Highlight. Our new benchmark aims to evaluate how well models can localize an event given a multimodal semantic query that consists of a reference image, which depicts the event, and a refinement text to adjust the images' semantics. To systematically benchmark model performance, we include 4 styles of reference images and 5 types of refinement texts, allowing us to explore model performance across different domains. We propose 3 adaptation methods that tailor existing models to our new setting and evaluate 10 SOTA models, ranging from specialized to large-scale foundation models. We believe this benchmark is an initial step toward investigating multimodal queries in video event localization. Our project can be found at httos://icq-benchmark.github.io/.

• Curated by: Gengyuan Zhang, Mang Ling Ada Fok
• Funded by [optional]:
  • Munich Center of Machine Learning
  • LMU Munich
• Language(s) (NLP): English
• License: Creative Commons Attribution Non Commercial Share Alike 4.0

Dataset Sources [optional]

• Repository: ICQ-benchmark

Uses

The dataset shall only be used for research purposes.

-->

Dataset Structure

• icq_highlight_release.jsonl: annotation file
• val_style_*.zip: reference images

Dataset Creation

Curation Rationale

We introduce our new evaluation dataset, ICQ-Highlight, as a testbed for Video Event Localization with Mulitmodal Queries. This dataset is built upon the validation set of QVHighlight, a popular natural-language query-based video localization dataset. For each original query in QVHighlight, we construct multimodal semantic queries that incorporate reference images paired with refinement texts. Considering the reference image style distribution discussed earlier, \dataset features 4 variants based on different image styles.

Source Data

Data Collection and Processing

We generate reference images based on the original natural language queries and refinement texts using a suite of state-of-the-art Text-to-Image models. Image generation can suffer from significant imperfections in terms of semantic consistency and content safety. To address these issues, we implement a quality check in two stages: (1) We calculate the semantic similarity between the generated images and the text queries using BLIP2~\cite{li2023blip} encoders, eliminating samples that score lower than 0.2; (2) We perform human sanity check to replace images that are: i) semantically misaligned with the text, ii) mismatched with the required reference image style, iii) containing sensitive or unpleasant content (\eg, violent, racial, sexual content), counterintuitive elements, or obvious generation artifacts.

Who are the source data producers?

• We build on the annotation files provided by QVHighlighs
• We generate reference images based on the original natural language queries and refinement texts using a suite of state-of-the-art Text-to-Image models, including DALL-E-2 and Stable Diffusion.

Annotations [optional]

Annotation process

We emphasize the meticulous crowd-sourced data curation and annotation effort applied to QVHighlight for 2 main reasons: (1) To introduce refinement texts, we purposefully modify the original semantics of text queries in QVHighlight to generate queries that are similar yet subtly different; (2) Given that the original queries in QVHighlight can be too simple and ambiguous to generate reasonable reference images, we add necessary annotations to ensure that the generated image queries are more relevant to the original video semantics.

We employed human annotators to annotate and modify the natural language queries. Each query is annotated and reviewed by different annotators to ensure consistency.

Personal and Sensitive Information

Despite manual filtering, there might be unpleasant/hallucinating synthesized images. Please contact us if you find any image offensive.

Bias, Risks, and Limitations

The dataset is synthesized and could include sensitive/unpleasant/hallucinating images.

Recommendations

Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations.

Citation [optional]

BibTeX:

[TBD]

Glossary [optional]

• cinematic: cinematic style images
• cartoon: cartoon style images
• scribble: scribble style images
• realistic: realistic style images

Dataset Card Authors [optional]

• Gengyuan Zhang
• Mang Ling Ada Fok

Dataset Card Contact

email: Gengyuan Zhang