Skip to main content
PHATPHUM.ME
All projects

Academic·2025

Fashion Object Detection

CS271 AI Fundamentals: Object Detection for Automated Product Tagging

Fashion Object Detection
Role
AI Engineer
Team
Group Project
Year
2025

Key figures

3
Tags extracted per image
YOLO
Architecture
GPU cloud
Trained on
Roboflow
Data prepared with

A group project in CS271 (AI Fundamentals) focused on building a computer vision system that detects and extracts clothing attributes from images to automate product tagging for e-commerce. The project emphasizes practical tags for apparel—such as color, clothing type, and patterns—to reduce manual labeling effort and improve consistency across large product image catalogs.

The problem

  1. 01Manual product tagging is time-consuming and error-prone, leading to inconsistent metadata.
  2. 02Large product image catalogs require automated extraction of key attributes for search and categorization.
  3. 03The solution needs a fast, practical detection approach suitable for real-world e-commerce workflows.

The approach

  1. 01Owned dataset preparation and image labeling in Roboflow, defining target classes/attributes (e.g., color, clothing type, pattern) and ensuring consistent annotations.
  2. 02Adopted YOLO for object detection due to its speed and practicality for inference-oriented use cases.
  3. 03Managed training infrastructure on GPU Cloud (Runpod.io) to enable efficient training and iterative experimentation.
  4. 04Established a training and evaluation workflow, iterating on key settings to better fit real-world image variability.

The result

  1. 01Delivered a prototype that detects clothing items and extracts key attributes for automated tagging (e.g., color, type, patterns).
  2. 02Reduced manual tagging effort while improving metadata consistency for e-commerce inventory search and management.
  3. 03Produced a repeatable pipeline—from dataset preparation to training and inference—that can be extended with additional classes or attributes.

Retrospective

Challenges

  • Defining clear, consistent labeling rules for attributes (such as colors and patterns) to ensure annotation quality.
  • Handling variability in lighting, angles, and image quality, which can impact attribute recognition performance.
  • Managing GPU-cloud training cycles efficiently to balance iteration speed, stability, and resource cost.

Learnings

  • End-to-end dataset workflows for computer vision: class/attribute definition, labeling, and dataset versioning with Roboflow.
  • Practical YOLO training and evaluation for fast object detection in real-world scenarios.
  • Operating training infrastructure on GPU Cloud (Runpod.io) and structuring experiments for systematic improvements.
  • Translating an e-commerce goal (automated tagging) into concrete technical requirements for data and model design.

Interested in work like this?

Tell me about your brief — I reply within 24 hours.

Start a conversation