Unified-IO 2: Scaling Autoregressive Multimodal Models with Vision, Language, Audio, and Action

Keyphrases

• Audio 100%
• Vision-language 100%
• Multimodal Model 100%
• Text-to-image 66%
• Bounding Box 33%
• Unified Model 33%
• Generated Image 33%
• Training Model 33%
• Image Understanding 33%
• Robotic Manipulation 33%
• Art Performance 33%
• Natural Language Understanding 33%
• Training Corpus 33%
• Diverse Sources 33%
• Semantic Space 33%
• Output Image 33%
• Image Generation 33%
• Token-based 33%
• Shared Semantics 33%
• Video Understanding 33%
• Denoiser 33%
• Transformer Model 33%
• Multimodal Pre-training 33%
• Encoder-Decoder Transformer 33%
• Multisensory Instruction 33%

Computer Science

• Research Community 100%
• Strongest Result 100%
• Training Model 100%
• Image Understanding 100%
• Art Performance 100%
• Image Synthesis 100%
• Natural-Language Understanding 100%