对抗损失
在实践中进行最小—最大训练
对抗损失 是 CoddyKit 上的免费 Deep Learning Academy 课时。 这是第 2 节课，共 4 节。 你可以在下方免费阅读本课时的完整内容 — 然后在浏览器中使用内置代码编辑器和全天候 AI 导师进行实践。 这是 Deep Learning Academy 学习路径的一部分，你的进度在网页和 CoddyKit 应用中同步。 Deep Learning Academy 课程共包含 4 节课。
本课时的部分内容尚未翻译，以英文显示。
Loss Drives Both Players
Every GAN player learns from a loss that scores how well it did. The two losses pull in opposite directions, which is what makes training adversarial.
The Min-Max Game
GAN training is a min-max game: the discriminator tries to maximize its score while the generator tries to minimize it. Neither fully wins.
Binary Cross-Entropy
Most GANs measure each prediction with binary cross-entropy, the standard loss for a real-versus-fake yes or no decision.
criterion = nn.BCELoss()Discriminator on Real Data
First the discriminator sees real images and should call them real. Its real loss pushes those outputs toward the label 1.
out_real = D(real)
loss_real = criterion(out_real, ones)Discriminator on Fakes
Next it judges generated images and should call them fake. The fake loss pushes those outputs toward the label 0.
out_fake = D(fake.detach())
loss_fake = criterion(out_fake, zeros)Why detach the Fakes
When updating the discriminator you call detach on the fakes so gradients do not flow back into the generator on this step.
fake.detach()The Generator's Loss
The generator wants its fakes labeled real, so its loss compares the discriminator's verdict on fakes against the label 1.
out = D(fake)
loss_g = criterion(out, ones)The Non-Saturating Trick
Telling the generator to maximize real-ness, instead of just minimizing fake-ness, gives stronger early gradients. This is the non-saturating loss.
Two Updates, One Round
Each training round does two updates: first the discriminator on its combined loss, then the generator on its own. Order and timing matter.
A Moving Target
Because both networks change every step, each one chases a moving target. This is why GAN losses bounce around instead of falling smoothly.
Losses Are Not Accuracy
In a GAN a falling loss does not mean better images. You judge quality by looking at samples, not by the loss number alone. 👀
Quick Check
One detail trips up many beginners.
Recap: Adversarial Loss
You saw the min-max game: cross-entropy guides both players, fakes are detached for the judge, and samples, not loss, reveal real progress. 🎯
用 AI 导师学习 Python — 免费
在浏览器中编写并运行真实代码，获得全天候 AI 导师的即时帮助，并在网页或应用中继续学习。
- 课程
- 30
- 课程
- 120
常见问题解答
「对抗损失」课时是免费的吗？
是的 — 「对抗损失」的完整文本可在网页上免费阅读。要进行交互式练习（内置代码编辑器和全天候 AI 导师）并解锁 Deep Learning Academy 课程的其余内容，请升级到 CoddyKit PRO。 Deep Learning Academy 课程共包含 4 节课。
「对抗损失」这节课中我会学到什么？
在实践中进行最小—最大训练 你通过在浏览器中直接运行的动手代码来练习 Deep Learning Academy，全天候 AI 导师会在你学习这节课的过程中回答你的问题。
学习 Deep Learning Academy 需要有经验吗？
无需任何先前经验。CoddyKit 上的 Deep Learning Academy 课程适合初学者到高级学习者，你可以从这里开始或从头开始，按照自己的节奏学习。 这是第 2 节课，共 4 节。
「对抗损失」课时需要多长时间？
大多数 CoddyKit 课程大约需要 5–10 分钟。每节课都很精短且互动，所以你能稳步进步，并在网页和应用中从离开的地方继续。
我能在这节 Deep Learning Academy 课中编写并运行代码吗？
能。每节 Deep Learning Academy 课都包含内置代码编辑器，你可以在浏览器中直接编写并运行真实代码，并获得即时 AI 反馈 — 无需本地设置。