Age Model

This repository documents iterative experiments on facial age transformation using a pretrained GAN-based generator.
The project focuses on manipulating latent representations to control age-related attributes and examining how lightweight model optimization techniques affect visual quality.

Experiments were conducted in an interactive workflow, with repeated execution, parameter adjustment, and qualitative evaluation of intermediate results.

---

Project Motivation

The goal of this project is to explore age progression and regression through latent space manipulation while preserving identity-related features.
Rather than producing a single final output, the emphasis is placed on iterative experimentation and comparative analysis across different model configurations.

---

Experimental Results

1. Baseline Age Transformation

The baseline experiment applies latent age edits to a pretrained generator without model optimization.
The figure below shows the original input alongside younger and older variations generated from the same latent representation.

---

2. Quantization Experiments

To evaluate efficiency trade-offs, the generator was quantized and the same age manipulation procedure was repeated.
Results are compared qualitatively against the baseline to observe changes in texture, smoothness, and age consistency.

---

3. Pruning Experiments

Structured pruning was applied to reduce model capacity while retaining core functionality.
The resulting images highlight the impact of parameter removal on facial detail and age-related features.

---

Observations and Insights

• Latent-based age manipulation produces consistent semantic age changes across all configurations.
• Quantization introduces minor smoothing effects but largely preserves age directionality.
• Pruning has a more noticeable effect on fine-grained textures, especially in extreme age shifts.
• Iterative visual inspection was critical for selecting reasonable age offsets and optimization levels.

---

Tech Stack

• Python
• PyTorch
• Pretrained GAN-based generator
• GPU-accelerated execution (Colab-style environment)

---

Repository Contents

• Agemodel_colab_code.py — main experimental script
• images/
  • baseline_results.png
  • quantized_results.png
  • pruned_results.png
• README.md — experiment documentation

---

Notes

This repository represents an exploratory, experiment-driven project intended for portfolio development.
The focus is on understanding model behavior through repeated execution and qualitative evaluation rather than producing a production-ready system.