引導式測驗 · 中等 等級
Imagen 2 和獎勵調整擴散測驗
將記錄的 Imagen 2 功能與一般獎勵調整研究分開,並檢視編輯工具、浮水印和擴散控制。
問題 1 的 8
Imagen 2 使用什麼核心生成技術?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- Imagen 2 使用什麼核心生成技術?
- 將隨機雜訊降噪到影像中的擴散
- 自迴歸標記預測
- 遺傳演算法
- 基於規則的向量繪圖
- 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
- 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
- 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
- In supported Imagen 2 workflows, what does SynthID do to generated images?
- Embedding an invisible watermark to help identify AI-generated images
- 提高解析度
- 翻譯提示
- 壓縮檔案
- 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
- 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
- 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