Computer Vision · Research · 2026
Vision Transformer Evaluation Framework
Problem
Comparing Vision Transformer and CNN variants needed a clean evaluation flow and a presentation format that made the tradeoffs easy to read.
Contributions
- Built a repeatable benchmarking workflow for ViT and CNN model comparisons.
- Analyzed performance tradeoffs across training and evaluation configurations.
- Packaged the results into a poster-first presentation format.
Architecture
- Python training pipeline → PyTorch experiments → evaluation summaries
- Shared protocol keeps model comparisons consistent across runs
- Poster layout communicates methodology, metrics, and findings
Stack
PythonPyTorchComputer VisionViTCNNs