FashionQ: An AI-Driven Creativity Support Tool for Facilitating Ideation in Fashion Design

Title of the Paper

FashionQ: An AI-Driven Creativity Support Tool for Facilitating Ideation in Fashion Design

Paper Information

• Domain: Application of Artificial Intelligence and Creativity Support Tools (CST) in ideation for fashion design
• Keywords: Creativity support tools, AI utilization, fashion design, ideation process, cognitive operations

Research Background and Problem

• What problems or challenges did the authors identify?

  • Ideation in fashion design is often influenced by design fixation, which hinders designers from expanding or generating new ideas.
  • There is a lack of quantitative and structured methods for defining brand styles, trend analysis, and design directions. Design work heavily relies on subjective experience and intuition, which may lead to inefficiency.
  • Human analysis struggles to meet the demands of processing multi-season and large-scale fashion data, often resulting in information overload for designers.

• Why is this problem important?

  • Ideation is a core element of success in fashion design. Supporting designers' divergent and convergent thinking can enhance their creativity and design outcomes.
  • With the development of big data and AI, exploring how these technologies can be applied to the design field to improve efficiency and quality is a significant research direction.

• Research Motivation and Related Work:

  • Current research has introduced AI into creativity support tools, but most focus on graphic or text analysis, lacking specific studies on the needs of fashion design.
  • The authors aim to develop an AI-driven creativity support tool to explore how it can support divergent and convergent thinking, particularly in addressing design fixation.

Solution

• What methods or solutions did the authors propose?

  • Developed an AI-driven creativity support tool called "FashionQ," which includes three interactive visualization interfaces (StyleQ, TrendQ, and MergeQ) to support cognitive operations of extending, constraining, and blending.

• What are the innovative aspects of this solution?

  • Theoretical Innovation: Tool design guided by a cognitive operations framework for divergent and convergent thinking.
  • Technical Innovation: Utilized a large-scale fashion dataset of 302,772 images and deep learning models (e.g., RetinaNet and Non-Negative Matrix Factorization (NMF)) for fashion attribute detection, style clustering, and trend prediction.
  • Methodological Innovation: The tool quantitatively and visually assists designers in overcoming subjectivity and improving design efficiency.

• What are the implementation steps and key technologies used?

  1. Interviews and Needs Analysis: Conducted interviews with 10 professional fashion designers to analyze challenges and solutions in the ideation phase.
  2. Tool Development:
     • Style Detection Model: Used RetinaNet for attribute annotation to detect specific elements in fashion images, such as clothing types, colors, and materials.
     • Style Clustering Model: Applied Non-Negative Matrix Factorization (NMF) to categorize fashion images into 25 style categories.
     • Trend Prediction: Improved ARIMA time-series model to predict style popularity trends.
     • Three Modules:
       • StyleQ: Provides style clustering results based on attributes, extending designers' understanding of styles.
       • TrendQ: Displays temporal trends of specific styles, supporting convergent thinking.
       • MergeQ: Generates combinations of two style attributes, supporting the creation of new design directions.
  3. User Research and Tool Validation: Evaluated FashionQ's effectiveness through user experiments and interviews.

Research Outcomes

• What specific outcomes were achieved?

  • User Experiments: Verified through experiments with 10 fashion design professionals that FashionQ significantly supports ideation tasks, particularly excelling in divergent thinking (StyleQ module) and convergent thinking (TrendQ module).
  • Tool Advantages:
    • Simplified complex data analysis tasks, such as summarizing design trends across multiple seasons.
    • Provided quantitative support to overcome limitations of intuitive analysis by designers.

• What advantages does it have compared to existing solutions?

  • Integrated extending, constraining, and blending cognitive operations to comprehensively support different stages of ideation.
  • Enhanced the logic and intuitiveness of the design process through quantitative data analysis and visualization results.
  • Specifically developed for the fashion design domain, making it more relevant than previously generalized AI methods.

• What are the experimental or evaluation results?

  1. User Feedback:
     • StyleQ (supporting divergent thinking) and TrendQ (supporting convergent thinking) received high satisfaction scores (average ratings of 5.4 and 5.8 / 7, respectively).
     • MergeQ (supporting blending) received positive feedback for assisting in new design combinations but scored slightly lower for generating highly innovative design directions (4.1/7).
  2. Confidence Comparison:
     • Participants reported significantly higher confidence in results when using FashionQ compared to traditional design methods.
  3. Interview Analysis:
     • Reduced the time designers spent on repetitive manual analysis.
     • The comprehensiveness of data scale and trend analysis was highly trusted by designers.
     • Provided an AI-supported "design collaborator" perspective.

• Limitations and Future Directions:

  • Limitations:
    • The dataset annotation process is time-consuming and requires domain expertise.
    • Some designers expressed reservations about the transparency of model predictions and the rationale behind style clustering numbers.
    • User research was limited to experimental environments, lacking long-term validation in real-world design contexts.
  • Future Directions:
    1. Improve AI interpretability and controllability, such as allowing users to adjust attribute weights to optimize results.
    2. Extend FashionQ to the implementation phase of creative concepts, exploring rapid prototyping tools.
    3. Apply the FashionQ framework across other design domains, such as product design and interior design.
    4. Incorporate diverse datasets, such as e-commerce sales data and social media feedback information.