Imagen

• better top-1 accuracy on ImageNet than EfficientNet at similar latency
• Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding
• text-to-image diffusion model
• large Transformer language models in understanding text and hinges on the strength of diffusion models in high-fidelity image generation
• Imagen produces $1024\times 1024$ samples with unprecedented photorealism and alignment with text
• generic large language models (e.g. T5), pretrained on text-only corpora, are surprisingly effective at encoding text for image synthesis
• increasing the size of the language model in Imagen boosts both sample fidelity and image-text alignment much more than increasing the size of the image diffusion model
• FID score
• COCO