引導式測驗 · 中等 等級
Google 影像測驗
查看 Imagen 如何調節文字的擴散、改善影像細節以及在生成過程中使用指導和閾值。
問題 1 的 8
Google Imagen 是基於什麼類型的生成模型建構的?
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。
- Google Imagen 是基於什麼類型的生成模型建構的?
- 擴散模型
- 生成對抗網路(GAN)
- 僅變分自動編碼器
- 基於規則的渲染器
- Imagen 原始論文的一個重要發現是,縮放哪個組件最能改善結果?
- 影像擴散網絡
- 凍結文字編碼器(語言模型)
- 訓練批量大小
- 顏色通道數
- Which improvement did Google announce for Imagen 4 in the Gemini app?
- 將檔案另存為 PNG
- 產生音訊
- Improved text rendering in generated images
- 壓縮影像
- Imagen 的原始架構使用級聯,首先以什麼解析度產生影像?
- 1024x1024
- 512x512
- 4096x4096
- 64x64
- What did Google DeepMind say SynthID does for images generated with Imagen on Vertex AI?
- Embeds an imperceptible watermark in image pixels to help identify likely Imagen output
- Converts all images to a new file format
- Provides a GPU chip for image training
- Translates the prompt language
- 開發人員可以透過哪個 Google 平台存取 Imagen 以用於自己的應用程式?
- Google 表
- 頂點人工智慧
- Google 地圖
- Chrome 開發工具
- What issue does dynamic thresholding address in Imagen's sampling process?
- The model having too few text tokens in its prompt
- The training set containing too many image resolutions
- Pixel values becoming oversaturated when classifier-free guidance is high
- The tokenizer splitting a word into subwords
- What does classifier-free guidance do during Imagen sampling?
- Selects a class label after the image is complete
- Upscales the image without conditioning on its prompt
- Replaces the noise-prediction model with a fixed image lookup
- Combines conditional and unconditioned predictions to steer generation toward the prompt