To address this, ensure the conditioning input is fed correctly into both the generator and discriminator, and consider using techniques like label smoothing or improved loss functions.
Here is the code snippet you can refer to:
In the above code, we are using the following key points:
- Conditioning: Ensure labels are correctly concatenated with noise (generator) or images (discriminator).
- Loss: Use binary cross-entropy to distinguish real/fake samples with conditioning.
- Stability: Apply label smoothing or feature matching for stable training.
Hence, by referring to the above, you can solve your problem.