How duongtruongbinh/improve_vivqa-x_baseline works
Project type and purpose
• From the file‐tree and naming conventions (“src/inference/qwenvl.py”, “src/evaluation/eval.py”, “notebooks/test.ipynb”, “results/ViVQA-X_test_…json”) this is an ML evaluation pipeline for a vision-language model (Qwen2.5-VL-7B). Its main purpose is to take a set of vision-question inputs, run inference through the VL model, and compute evaluation metrics.
File structure
• Top-level:
• .gitignore, README.md
• requirements.txt (Python deps)
• configs/ (placeholder for configuration files)
• docker/ (Dockerfile or related scripts)
• notebooks/ (interactive analysis)
• scripts/ (utility or orchestration scripts)
• src/inference (model‐inference code)
• src/evaluation (evaluation code + stored evaluation_scores.json)
• results/ (output JSONs from inference/evaluation)
• tests/ (unit or integration tests)
• Patterns & architectural hints:
• Separation between inference and evaluation modules
• Use of Docker for environment reproducibility
• Configs directory suggests configurable hyperparameters or dataset paths
• Notebooks for exploratory data analysis / result visualization
• Scripts for glue logic (e.g., kicking off a full pipeline)
README
• (Empty in this case—rely entirely on folder names and code responsibilities.)