CtrlVTON paper introduces controllable virtual try-on via VIP-SAM segmentation
A new arXiv paper proposes CtrlVTON, a virtual try-on system that gives users control over garment fit, style, and placement. It introduces VIP-SAM for instance-level garment segmentation from flatlay images.
Entities
Related
Virtual try-on Chrome extension for any clothing store
A Chrome extension that lets users upload a photo of themselves and see a photorealistic render of any garment from any online store on their body. It solves the guesswork of online clothes shopping, reducing returns and improving fit confidence. The extension uses photorealistic rendering (likely AI-based) to generate the try-on image in about 10 seconds, all processed locally on the user's device.
WearWhat: AI wardrobe app that catalogs clothes and suggests outfits
WearWhat is a mobile app that lets users snap photos of their clothes, uses AI to remove backgrounds and auto-tag fabric/color/style, creating a searchable digital closet. It also offers AI try-on to visualize outfits and generates daily outfit ideas using only owned items. The app solves the problem of having a full closet but feeling like there's nothing to wear.
Activation-Guided GCG attacks target refusal direction in LLMs
A new arxiv paper introduces Activation-Guided GCG, an adversarial attack that optimizes suffixes by directly targeting a model's internal refusal direction in activation space, rather than output-based objectives. The work probes the geometry of safety representations, showing that refusal behavior is mediated by low-dimensional directions that can be suppressed via optimization.
LLM for EDA in Front-End Design: Challenges and Opportunities
A new arXiv paper surveys the potential of large language models (LLMs) in front-end electronic design automation (EDA), covering HDL generation, testbench construction, and design space exploration. It highlights agentic AI systems like OpenClaw as a roadmap for next-generation EDA, addressing growing chip complexity and time-to-market pressures.
Researchers identify asymmetric generalization problem in LLM unlearning benchmarks
A new arXiv paper argues that existing machine unlearning benchmarks for LLMs suffer from under-forgetting and over-forgetting due to an asymmetric generalization problem. The authors propose that evaluation must cover diverse query formulations of target facts to reliably measure knowledge removal while preserving unrelated capabilities.