Skip to main content

VAE RGB Importance

1 min

What I worked on

Looked into how RGB channels are handled during VAE training and how to visualize them in TensorBoard.

What I noticed

  • x was a flattened tensor representing RGB images
  • Using .view(64,108,3) reshaped the tensor incorrectly at times

”Aha” Moment

n/a

What still feels messy

Why reconstructed outputs show gray tones even when color data is present. Still not fully clear how Lightning and TensorBoard handle RGB channels internally.

Next step

Fix the visualization code by confirming tensor shape before logging and experiment with color normalization during training.