📗 -> 12/03/24: GAN
🎤 Vocab
❗ Unit and Larger Context
Small summary
✒️ -> Scratch Notes
Video GAN
In relevant situations, we don’t just want to deepfake random images. We want to condition our model on existing images, to ‘extend’ existing pictures/videos.
- Also a very difficult question. What should the model move? We as people now that the laws of physics apply, and that certain things do and don’t move (trains vs train tracks), but the model does not.
GAN has a latent space:
- The GAN serves as a transformation from noise to image
- Interestingly: you can do vector addition of noise inputs to get fun results
- Manipulating the noise that creates these images, not the images themselves
- Ties to Explainable-AI, and similar work being done on LLMs.
- For actual applications, you can imagine wanting to translate an image from daytime to night-time
- Image to Image Translation
Image-to-Image Translation with Conditional Adversarial Networks
- Looks intersting (creating a city view from satelite view but I don’t understand)
Other interesting things with GAN
- Text-to-Image Translation (text2image)
- Learning What and Where to Draw
- Semantic-Image-to-Photo Translation
- Generate New Human Poses
- Photos to Emojis
- Photograph Editing (+- bangs, smiling, etc)
- Image De-raining
- Photo Inpainting
- Face Image Inpainting
- Clothing Translation
🔗 -> Links
Resources
- Put useful links here
Connections
- Link all related words