Vizuara/robot-video-search

🤖 Robot Video Search Engine

Semantic search over 950K+ robot manipulation video clips using CLIP embeddings.

Features

• —Text Search: Describe what you're looking for in natural language
• —Image Search: Upload a robot/scene image to find similar clips
• —Random Browse: Explore the dataset randomly

Datasets

• —BridgeData V2: 42K clips from WidowX arm across 24 environments
• —Open X-Embodiment: 900K+ clips from 55 different robot types

Technical Details

• —Embeddings: CLIP ViT-L/14 (OpenAI), 768-dimensional
• —Captions: Generated by Qwen2.5-VL-7B-Instruct
• —Search: Cosine similarity over normalized embeddings
• —Deduplication: FAISS-based, threshold 0.95

Built by Vizuara as part of the World Foundation Model data curation pipeline.