STANLEY COLLEGE OF ENGINEERING AND TECHNOLGY FOR WOMEN (AUTONOMOUS)
Abids. Hyderabad – 500 001
(Affiliated to Osmania University & Approved by AICTE)
(All eligible UGC courses are accredited by NBA & NAAC with ‘A’ Grade)
Department of Computer Science and Engineering
Academic Year: 2025-2026
B.E Major Project Phase-I
Title :Virtual AI Stylist and Outfit Recommender
PRESENTED BY
Guide Name- Mrs Swathi Sree
Vishwanath Nikitha- 160622733186
Assistant Professor
Nausheen Tarannum- 160622733167
Department of CSE
Pandirla Nishitha- 160622733189
Review – I Suggestions
• Finalize the domain properly.
• Identify a clear real-world problem.
• Write a crisp problem statement.
• Prepare 2–3 possible project titles and choose the best one.
Review – II Suggestions
• Write a simple and meaningful abstract.
• Add background explanation: introduction, need, challenges.
• Start reading good-quality recent research papers (last 5 years).
• Understand methodology, dataset, and tools used in papers.
Review – III Suggestions
• Prepare a PPT based on 5–8 research papers.
• Compare methods and highlight gaps.
• Select one suitable base paper for your project.
• Finalize title, problem statement, and abstract clearly.
Contents
❖Abstract
❖Introduction
❖ About the Project
❖ Problem Statement
❖ Objectives of the Project
❖ Scope of the Project
❖ Applications of the Project
❖Literature Survey
❖Research Gap
❖Proposed Methodology
❖Planning Of Work
❖References
Abstract
This project builds a Virtual Fashion Stylist using Machine Learning. The shopper uploads a
photo, and the system detects their face to identify their skin tone (fair, medium, or dark). The
user also selects the occasion like a wedding, party, casual outing, or formal event.
All outfits in the store are already labelled with their colours, styles, and suitable events. The
ML model matches the user’s skin tone and occasion with these outfit details. It then gives a
suitability score or a simple “recommended / not recommended” result.
This creates a personalized, quick, and smart shopping experience, helping users choose
outfits that look good on them and fit the event without needing complex fashion rules.
Introduction
Shoppers struggle to choose outfits that match their complexion and suit events. Existing systems
usually ignore these two factors together. Our project uses skin-tone detection and machine learning
to give personalised and event-appropriate outfit suggestions.
About the Project
Input: User photo + selected event .
Process: Skin-tone detection → outfit feature extraction → ML suitability scoring.
Output: Recommended / Not Recommended + ranked outfit list.
Problem Statement
Users cannot easily judge if an outfit suits their complexion or matches an event. Current tools lack
combined complexion + occasion analysis.
Objectives
• Build a skin-tone classifier
• Create an outfit dataset with colour, pattern, event labels
• Train ML model to predict suitability
• Provide simple, clear recommendations
Scope of the Project
Recommends outfits based on skin tone and event type.
Uses ML models for accurate, personalised suggestions.
Includes a labelled outfit database (colours, style, occasion).
Can be integrated into websites, apps, or retail systems.
Scalable with options to add advanced styling features.
Applications of the Project
Online shopping platforms for personalised outfit suggestions.
Retail stores using smart mirrors/kiosks.
Personal styling and fashion assistant apps.
Virtual try-on and event-based outfit selection.
Tools to improve customer engagement in fashion retail.
Literature Survey
1: Author: Rucong Xu, Jianfeng Wang, Yun Li Title: Heterogeneous-Grained Multi-Modal Graph Network for
Outfit Recommendation Journal: IEEE Transactions on Emerging Topics in Computational Intelligence Year of
Publication: 2024
Problem Taken Improving outfit recommendation by understanding how clothing items match
at both overall and detailed levels.
Applied Algorithm/Methodology Uses a multi-modal graph network combining image, text and category features
to score outfit compatibility.
Data Set Polyvore datasets containing outfit images, text descriptions and labels.
Performance Measures Compatibility accuracy, retrieval accuracy, and fill-in-the-blank scores.
Improvements Gives more accurate outfit matches by learning from multiple types of features.
Limitations Does not consider skin tone, user preferences, or event suitability.
Literature Survey
2 : Author: Su, X., Duan, J., et al Title: Developing an AI-Based Automated Fashion Design SystemJournal:
IJARSCT Year of Publication: 2024
Problem Taken Reducing manual effort in fashion design by automating design analysis and
pattern prediction.
Applied Algorithm/Methodology ML-based pattern classification, image segmentation, feature extraction for
clothes.
Data Set Fashion design images with labelled patterns and colour information.
Performance Measures Prediction accuracy, design identification rate, processing speed.
Improvements Improves consistency in fashion design, reduces manual work, increases design
accuracy.
Limitations Limited to predefined fashion patterns and simple designs; lacks personalization
for users
Literature Survey
3: Author: Choi, [Link]: A Mixed Reality Virtual Clothes Try-On System Journal: Not SpecifiedYear of Publication:
Not Mentioned
Problem Taken Allowing users to virtually try outfits without physically wearing them.
Applied Algorithm/Methodology 3D clothing modelling, augmented reality overlay, body tracking.
Data Set 3D clothing models and user body images.
Performance Measures Rendering quality, try-on accuracy, user interaction feedback.
Improvements Enables realistic outfit visualization before buying clothes.
Limitations Requires high processing power and special devices; not purely ML-based.
Literature Survey
4 : Author:Yuan, M., et al. Title: Dress-Up: A Deep Unique Personalized Fashion Recommender Journal: Stanford
CS231n Year of Publication: 2024
Problem Taken Lack of personalized outfit recommendations based on a user’s style.
Applied Algorithm/Methodology Uses CNNs to learn user preferences and match them with suitable outfits.
Data Set User outfit images and labelled style preferences.
Performance Measures Recommendation accuracy and personalization quality.
Improvements Provides better personalized results using deep learning.
Limitations High computational cost and requires large datasets.
Literature Survey
5 :Author:Chen, W., Huang, P., Xu, J., et al. Title: Personalized Outfit Generation Journal: arXiv Year of
Publication: 2019
Problem Taken Generating outfits where all items match well visually.
Applied Algorithm/Methodology Deep learning-based compatibility scoring and outfit combination generation.
Data Set Fashion datasets with labelled compatible outfits.
Performance Measures Compatibility accuracy and generation quality.
Improvements Improves the creation of visually matching outfit sets.
Limitations Does not consider user complexion or event needs.
Literature Survey
6 :Author: Sagar, D., Garg, J., Kansal, P., Bhalla, S., Shah, R. R., & Yu, Y. Title: PAI-BPR: Personalized Outfit
Recommendation Journal: arXiv Year of Publication: 2020
Problem Taken Increasing interpretability in outfit recommendations.
Applied Algorithm/Methodology Uses Bayesian Personalized Ranking with attribute attention.
Data Set Fashion images with labelled colours and style attributes.
Performance Measures Ranking accuracy and interpretability score.
Improvements Explains why a recommendation was made.
Limitations Does not handle skin-tone or event-based matching.
Literature Survey
7 :Author: Li, Y., Fu, X., & Du, Y. Title: Prediction of Outfit Compatibility Based on Weighted Multimodal
Fusion Journal: —Year of Publication: 2024
Problem Taken Checking how well two or more clothing items match each other.
Applied Algorithm/Methodology Uses weighted fusion of image and metadata features
Data Set Fashion images with compatibility information.
Performance Measures Compatibility prediction accuracy.
Improvements More accurate matching by combining multiple feature types.
Limitations Does not personalise results for specific users.
Literature Survey
8 : Author: Dixit, S., Dagde, R. et al Title: Enhancing Outfit Recommendation with CNN-kNN Hybrid Model
Journal: IJRASET Year of Publication: 2024
Problem Taken Improving the accuracy of outfit recognition and matching.
Applied Algorithm/Methodology CNN for extracting features + kNN for nearest outfit matching.
Data Set Fashion datasets labelled by colours and patterns.
Performance Measures Recognition accuracy and matching precision.
Improvements Gives fast and reliable outfit suggestions
Limitations Lacks personalization and deeper compatibility analysis.
Literature Survey
9 : Author: Xueping Su, Duan Title: Personalized Clothing Recommendation Using 4-Season Color System
Year of Publication: 2024
Problem Taken Matching clothing colours with user skin tone and facial features.
Applied Algorithm/Methodology Uses MobileNetV3 to detect skin and colour tones and map them to seasonal
palettes.
Data Set Face images with labelled seasonal colour categories.
Performance Measures Skin-tone classification accuracy.
Improvements Gives better colour matching for complexion.
Limitations Does not consider event type or personal style.
Literature Survey
10 : Author: Hongbo Ma, Fei Shen, Hongbin Xu, Xiaoce Wang Title: Virtual Try-On with Diffusion Models
Journal: arXiv Year of Publication: 2025
Problem Taken Making virtual try-on images more natural and realistic.
Applied Algorithm/Methodology Uses diffusion models to generate realistic clothing overlays.
Data Set User photos paired with garment images.
Performance Measures Visual realism, quality score, user feedback.
Improvements Creates highly realistic virtual try-on results.
Limitations Very computationally heavy and not suitable for simple ML systems
Paper No. Research Gap
1 Needs a lot of data and strong computers; not suitable for simple or real-time use.
2 Designs clothes but does not recommend outfits based on user needs.
3 Shows virtual try-on but does not give smart or personalised suggestions.
4 Works well but may not handle many users or limited data.
5 Creates outfits but doesn’t check if they are practical for real people/events.
6 Struggles with new users and doesn’t use skin tone or colour matching.
7 Combines features but doesn’t show best way to mix them or handle bad data.
8 Slow with big datasets and doesn’t personalise recommendations well.
9 Depends too much on perfect skin tone detection, which is not always accurate.
Makes realistic try-ons but doesn’t check if the outfit suits the person or event; needs strong
10
hardware.
PROPOSED METHODOLOGY
1. Image Acquisition- User uploads a clear face photo.
2. Face & Skin Extraction- Face detected → skin region isolated using colour spaces.
3. Skin Tone Classification -ML classifier (SVM/Random Forest) predicts skin tone category
(Fair/Medium/Dark).
4. Event Input- User selects an occasion such as wedding, party, formal or casual.
5. Outfit Dataset Preparation -Outfits labelled with colours, styles, and suitable event categories.
6. Suitability Prediction - ML model checks compatibility between skin tone, outfit colours and event
suitability. Outputs suitability score or “Recommended / Not Recommended.”
7. Recommendation Generation - Top outfits displayed with colour match level and event relevance.
8. User Interface- Simple UI for uploading images, selecting events, and viewing suggestions.
Planning of the Work
Phase 1: Requirement Analysis - Identify user needs, finalise skin-tone categories and event
types.
Phase 2: Dataset Preparation - Collect and label outfit images and sample user photos.
Phase 3: Image Processing - Implement face detection and skin-pixel extraction.
Phase 4: ML Model Development - Train skin-tone classifier and suitability prediction model.
Phase 5: Module Integration - Combine skin tone, outfit dataset, event selection and prediction
model.
Phase 6: UI Design - Create a simple interface for input and recommendations.
Phase 7: Testing & Validation - Test with multiple images and improve accuracy.
Phase 8: Documentation & Deployment - Prepare report and deploy prototype
for demonstration.
References
[1] Xu, R., Wang, J., & Li, Y. (2024). Heterogeneous-Grained Multi-Modal Graph Network for Outfit Recommendation. IEEE Transactions on Emerging Topics
in Computational [Link]: [Link]
[2] Su, X., Duan, J., et al. (2024). Personalized Clothing Recommendation Fusing the 4-Season Color System and Users’ Biological [Link]:
[Link]
haracteristics
[3] Choi, W. (2023). Developing an AI-Based Automated Fashion Design [Link]: [Link]
[4] Yuan, M., et al. (2013). A Mixed Reality Virtual Clothes Try-On [Link]:
[Link]
[5] Chen, W., Huang, P., Xu, J., et al. (2019). POG: Personalized Outfit Generation for Fashion [Link]: [Link]
[6] Sagar, D., Garg, J., Kansal, P., Bhalla, S., Shah, R. R., & Yu, Y. (2020). PAI-BPR: Personalized Outfit Recommendation Scheme with Attribute-wise
[Link]: [Link]
[7] Li, Y., Fu, X., & Du, Y. (2022). Prediction of Outfit Compatibility Based on Weighted Multimodal [Link]:
[Link]
[8] Dixit, S., Dagde, R. et al. (2024). Enhancing Outfit Recommendation with a CNN-kNN Hybrid Model and Digital Wardrobe [Link]:
[Link]
[9] Su, X., Duan, J., et al. (2024). Personalized Clothing Recommendation Using the 4-Season Color [Link]:
[Link]
haracteristics
[10] Xu, Y., et al. (2024/2025). Virtual Try-On / Diffusion-Based Fashion Generation (DiFashion).
