Slimmable Compressive Autoencoders (SlimCAEs)

Slimmable Compressive Autoencoders (SlimCAEs) is a neural image compression model that dynamically adjusts the network width to achieve variable compression rates. The model utilizes slimmable Generalized Divisive Normalization (GDN) layers and a λ-scheduling training algorithm to optimize memory usage, computation, and latency while maintaining high compression performance.

Features

• Dynamic Network Width: Adjusts the network's width to achieve different compression rates.
• Slimmable GDN Layers: Uses slimmable GDN layers for flexible image compression.
• λ-Scheduling Algorithm: Optimizes the model's training for better performance with varying compression levels.
• Optimized Memory & Computation: Reduces computational costs and memory usage while preserving compression quality.

Prerequisites

• Python 3.x
• TensorFlow or PyTorch (depending on your implementation)
• CUDA-enabled GPU for training (optional but recommended)