You have a dataset of face images at 128×128 resolution, som…

You have a dataset of face images at 128×128 resolution, some are severely noisy (grainy camera shots). You want to classify each image into one of five expressions: happy, sad, angry, surprised, neutral. You decide to build: Autoencoder (AE) for denoising. CNN that classifies the AE’s output. GAN for data augmentation—generating extra images in each expression category. After some early success, you suspect domain mismatch and overfitting. Let’s see what goes wrong. — Angry is the smallest class in the dataset. You generate GAN samples to augment. A post-hoc analysis shows some generated “angry” faces look more “cartoonish” or “mildly annoyed” than truly “angry.” Which statements about possible solutions are valid?

Now your dataset has short video clips of faces showing an e…

Now your dataset has short video clips of faces showing an expression transition (e.g., neutral → smile). Some clips are shot in low-light conditions. You attempt: GAN to brighten or color-correct frames, AE for further denoising or super-resolution, CNN (or 3D CNN) for expression classification across frames. After some usage, you realize certain frames come out “over-bright” or “washed out.” — XYZ