Тест, свързан с ръководство · Среден Ниво

Imagen 2 и Дифузионен тест с настройка на наградата

Отделете документираните възможности на Imagen 2 от общите изследвания за настройка на възнагражденията и прегледайте инструментите за редактиране, водните знаци и контролите за дифузия.

Свързани ръководни пътекиImagen 2 и Reward Tuned Diffusion
Въпрос 1 на 8

Каква основна генеративна техника използва Imagen 2?

Визуализирайте всеки въпрос8 въпроси

Отговорите и обясненията се появяват, докато отговаряте на теста.

  1. Каква основна генеративна техника използва Imagen 2?
    • Дифузия, която обезшумява случаен шум в изображение
    • Авторегресивно прогнозиране на токени
    • Генетични алгоритми
    • Векторно рисуване, базирано на правила
  2. In reward-tuned text-to-image research, what does a learned reward model do?
    • It removes the image diffusion process
    • It scores generated images according to patterns in human feedback
    • It converts images into shipping labels
    • It replaces the text prompt with random noise
  3. How is human-preference tuning for text-to-image models often compared with language-model alignment?
    • It is the same process as WordPiece tokenization
    • It uses only hand-written image rules
    • It is conceptually similar to reinforcement learning from human feedback (RLHF)
    • It removes all human judgments from training
  4. In the cited reward-weighted text-to-image method, what objective guides fine-tuning?
    • Minimizing image file size regardless of the prompt
    • Predicting the next text token without using image data
    • Maximizing the number of colors in every output
    • Maximizing likelihood weighted by a reward function trained on human feedback
  5. In supported Imagen 2 workflows, what does SynthID do to generated images?
    • Embedding an invisible watermark to help identify AI-generated images
    • Повишаване на резолюцията
    • Превеждане на подканата
    • Компресиране на файла
  6. Which Imagen 2 improvement did Google specifically report in its 2024 announcement?
    • Native generation of spoken audio
    • More realistic hands and faces and fewer distracting visual artifacts
    • A guarantee that every object count is exact
    • Replacing diffusion with autoregressive text prediction
  7. What tradeoff does classifier-free guidance control in diffusion generation?
    • Image width versus token vocabulary size
    • Watermark visibility versus font size
    • Prompt adherence versus sample diversity
    • Training data volume versus ship speed
  8. What did Google document for Imagen 2.0 on Vertex AI in 2024?
    • Real-time speech transcription
    • Automatic conversion of images into 3D CAD files
    • A guarantee of exact text rendering in every image
    • Inpainting and outpainting for photo editing