引導式測驗 · 中等 等級
Imagen 視訊級聯測驗
探索 Imagen Video 的分階段擴散管道,其中時間和空間超解析度模型逐步細化剪輯。
問題 1 的 8
Which design helps Imagen Video represent both spatial detail and changes across frames?
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。
- Which design helps Imagen Video represent both spatial detail and changes across frames?
- A Video U-Net that uses temporal self-attention in the base model
- A frozen spreadsheet encoder with no video operations
- A speech-only recurrent network
- A single still-image classifier applied once
- 時間超解析度級在級聯中扮演什麼角色?
- 提高像素解析度
- 在現有框架之間新增更多框架
- 翻譯提示
- 刪除音軌
- 什麼組件對 Imagen Video 中的文字提示進行編碼?
- 卷積影像分類器
- 語音辨識模型
- 凍結的 T5 語言模型
- 強化學習代理
- Why did Imagen Video use noise conditioning augmentation in its super-resolution cascade?
- Increase pixel resolution at the final spatial stage
- Make the prompt encoder generate extra text tokens
- Remove the need to condition the super-resolution models
- Reduce sensitivity to domain gaps between cascade stages, which also facilitates parallel training
- Imagen Video 特別展示了哪些功能?
- 渲染清晰的風格化文字和多樣化的藝術風格
- 電影製作時間
- 即時遊戲
- 僅編輯現有照片
- 整個 Imagen Video 管道的目標輸出大致是什麼?
- 純音訊剪輯
- 約 1280x768,每秒 24 幀
- 8K 120 fps
- 單一靜止影像
- Which diffusion parameterization did Imagen Video report using as a design choice?
- Only a nearest-neighbor color lookup
- A token-by-token autoregressive text decoder
- v-parameterization
- A fixed histogram equalization rule
- What was progressive distillation used for in Imagen Video?
- Adding subtitles after generation
- Replacing the prompt encoder with a database
- Converting generated video into handwritten text
- Faster sampling while retaining high-quality generated video