Skip to content

Comments

Lbt funcionando#25

Open
mmp819-max wants to merge 10 commits intoSistemas-Multimedia:mainfrom
aek676:LBT
Open

Lbt funcionando#25
mmp819-max wants to merge 10 commits intoSistemas-Multimedia:mainfrom
aek676:LBT

Conversation

@mmp819-max
Copy link

No description provided.

aek676 and others added 10 commits January 15, 2026 14:15
Add .devcontainer/devcontainer.json to provide a Python 3 dev
container (python:3.13.7-bookworm) with common-utils and python
features; installs system libs and runs pip to install requirements.
Add src/__pycache__/ to .gitignore. Add ipykernel and nbformat to
requirements.txt for Jupyter support.
- Introduced LBT class for spatial redundancy exploitation using 2D Discrete Cosine Transform.
- Implemented training, encoding, and decoding methods for LBT.
- Added command-line arguments for block size, color transform, and perceptual quantization.
- Updated requirements.txt to include 'torch' for PyTorch support.
…l problema. Con bloques grandes, la varianza puede ser muy pequeña, causando que 1/var explote y domine completamente el entrenamiento, haciendo que la transformación colapse a valores uniformes (gris).

Learning rate adaptativo: lr = min(1e-2, 1.0 / N) - Para bloques de 32 (N=1024) esto da lr≈0.001, evitando inestabilidad.

Gradient clipping - Previene explosión de gradientes con matrices grandes.

Ahora la transformación solo optimiza la reconstrucción, manteniendo las matrices cercanas a la identidad (que es la inicialización) y convergiendo correctamente.
…compression

- Introduced authorship section in the notebook.
- Added a theoretical explanation of the Learning Block Transform (LBT) including model architecture and encoding flow.
- Explained block size considerations and their impact on image processing.
- Detailed the training process of the LBT model, including loss functions and optimization strategies.
- Included information on quantization techniques and perceptual quantization.
- Added implementation details for the LBT model in a new Python file.
@dsa069
Copy link

dsa069 commented Feb 2, 2026

Buenas tardes. En nombre del grupo compuesto por:

  • Daniel Salas Alonso

  • José Carlos Ruíz Sánchez

  • Millán Millán Posadas

  • Anass El Jabiry Kaddir

Hacemos entrega de nuestro algoritmo LBT propuesto.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants