This repository explores the visual evolution of the luxury fashion house Dior using computer vision and CLIP-based embeddings. It demonstrates how to scrape, process, and analyse runway collections to reveal brand-specific design languages, creative director eras, and aesthetic trends over time. Dior was chosen as the main example because it features distinct creative transitions from John Galliano’s theatrical maximalism to Raf Simons’ architectural minimalism and Maria Grazia Chiuri’s feminist romanticism, making it an ideal case to study how design identity evolves through leadership changes. The framework can easily be adapted to analyse other fashion houses.
This end-to-end pipeline combines AI, fashion history, and data visualisation to map how visual styles evolve across decades and creative eras.
- Automated Runway Scraping (via Selenium + BeautifulSoup) (Source: https://www.firstview.com/)
- Structured Metadata Generation for Dior collections
- Visual Embedding Extraction using OpenAI’s CLIP
- t-SNE & PCA Style Maps to visualise aesthetic evolution
- Semantic Concept Scoring (“romantic”, “minimalist”, “avant-garde”…)
- Interactive Streamlit Dashboards for visual exploration
git clone https://github.com/axelsgit/fashion-visual-intelligence.git cd fashion-visual-intelligence 2. Install dependencies pip install -r requirements.txt 3. Run the end-to-end pipeline Pay attention to order of scripts.
image_scraper.py
create_metadata_csv.py
create_full_metadata_csv.py
extract_embeddings.py
visualise_evolution.py
compute_concept_similarity.py
streamlit run app_dashboard_dior.py
fashion-visual-intelligence/
│
├── image_scraper.py # Scrapes runway photos from FirstView
├── create_metadata_csv.py # Creates Dior base metadata CSVs
├── create_full_metadata_csv.py # Expands metadata with all image file paths
├── extract_embeddings.py # Extracts CLIP embeddings for all images
├── visualise_evolution.py # Performs PCA + t-SNE visual analysis
├── compute_concept_similarity.py # Computes similarity to style concepts
├── app_dashboard_dior.py # Streamlit dashboard for Dior
│
├── data/ # data created within scripts
│ ├── metadata_dior_womenswear.csv
│ ├── metadata_dior_womenswear_full.csv
│ ├── dior_concept_scores.csv
│ ├── dior_tsne_results.csv
│ ├── embeddings_metadata_dior.csv
│ ├── embeddings_clip_dior.npy
│ └── collection image folders
│
├── assets/ # screenshots
├── requirements.txt
└── README.md
Visual clusters showing shifts between creative eras and seasonal directions.
Preview of a few random looks from a chosen collection.
Dominant colour for each analysed collection over the decades.
Explore how AI understands aesthetic concepts like “romantic”, “minimalist”, or “avant-garde” in Dior’s collections. The model compares the runway image with the text description of that concept, highlighting which collections most embody each aesthetic.
This timeline shows how stylistic alignment with key concepts evolves over time. Peaks and dips in each line reflect the brand’s changing priorities. The shaded backgrounds indicate the tenure of each creative director.
1 Scraping:
Uses Selenium to load dynamic image galleries from FirstView runway archives.
2 Metadata Structuring:
Each collection is tagged with year, season, director, and creative era.
3 Feature Embedding:
CLIP (ViT-B/32) converts each image into a high-dimensional aesthetic vector.
4 Dimensionality Reduction:
PCA → t-SNE or PCA 3D visualises stylistic similarity and drift.
5 Concept Scoring:
CLIP measures how each look aligns with style descriptors like “romantic”, “futuristic”, or “minimalist”.
6 Dashboard Visualisation:
Streamlit dashboards let you explore evolution interactively by era, year, or concept, enables preview of looks by collection and aesthetic concept, and shows development of dominant colours and aesthetic language.
- John Galliano (1998–2011) - maximalist theatrical couture
- Raf Simons (2012–2015) - minimal architectural silhouettes
- Maria Grazia Chiuri (2016–2024) - feminist romantic modernism
The analysis revealed patterns that align with Dior’s well-known aesthetic DNA, a blend of romanticism, avant-garde experimentation, and timeless femininity.
Across decades, AI-based visual analysis captured the transition from John Galliano’s theatrical maximalism, through Raf Simons’ architectural minimalism, to Maria Grazia Chiuri’s poetic and feminist romanticism.
However, the CLIP-based trend chart shows that these shifts are not always reflected in the model’s outputs. While Raf Simons’ era is widely known for its minimalist approach, the AI detected only a weak similarity to the concept “minimalist.” This suggests that CLIP’s interpretation is based primarily on surface-level visual cues rather than the underlying conceptual or structural principles of minimalism.
These results highlight both the potential and the limitations of AI in fashion interpretation. Dior’s enduring appeal lies in how its creative directors continue to evolve the brand’s narrative language while maintaining a coherent emotional and stylistic core.
| Creative Director | Era | Aesthetic Core | Visual Traits | AI Concept Alignment (CLIP Similarity) |
|---|---|---|---|---|
| John Galliano | 1998–2011 | Theatrical Romanticism | Dramatic silhouettes, bold color, fantasy couture | Romantic ↑ Avant-garde ↑ |
| Raf Simons | 2012–2015 | Modern Minimalism | Architectural lines, purity of form | Minimalist (weak correlation) |
| Maria Grazia Chiuri | 2016–2024 | Feminist Romanticism | Soft draping, symbolism, wearable narrative | Romantic ↑ Futuristic ↑ |
- This project uses publicly accessible runway imagery from FirstView and similar fashion archives for non-commercial, educational, and research purposes only.
All images remain the property of their respective copyright holders.
Visual samples shown in this dashboard are low-resolution thumbnails used solely for academic illustration and AI-based visual analysis. - The project does not redistribute or claim ownership of any images and is not affiliated with or endorsed by Dior, FirstView, or any associated entities.
- All data and results are provided for academic visualisation and fashion analysis.
- If you are a rights holder and wish for specific content to be removed, please contact the repository maintainer for prompt removal.
👩💻 Author: Axel Heussner