知识卡片
前向过程提供监督样本
预先定义的损坏过程可从干净训练图像构造不同噪声水平的输入和预测目标,用于学习反向转换。可预测较干净图像、原图、噪声或转移分布参数,但每种目标都需与相应采样公式一致。训练对构造并不要求有人为生成图像逐张配标签。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 生成模型
原书出处 · 1 处
- · 查看原书依据
Now, just like autoregressive models, diffusion models train a neural net, fθ, to reverse this process. It can be trained via supervised learning on examples sequences of different images getting noisier and noisier. For example, as shown below in figure 32.15, the sequence of images of the bird getting noisier and noisier can be flipped in time and thereby provide a sequence of training examples
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第32章《生成模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c032.xhtml)
- 结论依据:摘录中关于前向过程提供监督样本的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:Now, just like autoregressive models, diffusion models train a neural net, fθ, to reverse this process. It can be trained via supervised learning on examples sequences of different images getting noisier and noisier. For example, as shown below in figure 32.15, the sequence of images of the bird getting noisier and noisier can be flipped in time and thereby provide a sequence of training examples