AI and Fashion- Product Retrieval with Multi-modally Generated Data
summary
Claims a working pipeline that turns an image + text query into a synthetic image used for visual search, built with FashionIQ dataset, ResNet pretrained embeddings and multimodal encoders, and a generator conditioned on concatenated image‑and‑text features. It augments conditional GANs (AttnGAN/StackGAN ideas) with a conditional GAN + triplet loss so generated images are closer to ground‑truth targets than random negatives, improving multimodal retrieval over earlier classification‑only baselines; work is promising but in progress.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:14Why product retrieval matters
- 3:23Visual search basics
- 4:49Siamese and triplet networks
- 7:18Visual-only search limits
- 8:41Multimodal retrieval with deep style
- 10:05Semi-supervised multimodal embeddings
- 11:54Production challenges and interpretability
- 12:59Generative models introduction
- 14:01Conditional GANs in fashion
- 15:30Text-to-image generation
- 17:40GANs with triplet loss
- 20:50Synthetic data features
- 21:11Summary and future research
- 21:33Q&A: Wardrobe revamp
- 22:16Q&A: Commercial plans
- 23:18Q&A: Future potential
related talks
34mComputer Vision and NLP for Multi-Task Fashion Attrb. Modeling
32mGenerative AI Text and Multimodal Embedding Models for Real World Use Cases
35mGenerative AI for Fraud Detection & Claims Analysis
22mHarnessing image and text data for product matching
26mMaximizing Multimodal: Exploring the search frontier of text-to-image models
30mGenerative AI will enable crowdsourced interactivity
33mMachine Learning in Visual Search
01h 26mBuilding an Image Search Engine with Generative Computer Vision Models