Multi_Image_MLLM

Cleaned project layout focused on the main workflow.

Main workflow

1. Generate or sample datasets in gen_data/.
2. Run evaluations with test/run_tests.sh.
3. Review model outputs in test/result/.
4. Run analysis scripts in analysis/.

Key directories

• gen_data/: dataset generation and sampling scripts.
• test/: evaluation runner and test logic.
• test/result/: generated evaluation outputs.
• analysis/: post-processing and statistics scripts.

Core entrypoints

• test/run_tests.sh: primary test runner.
• test/test.py: evaluation logic used by the runner.
• test/questions.py: question definitions and evaluation helpers.
• unified_mllm.py: unified model wrapper.

Notes

• Most one-off debug/example scripts were removed to keep the project focused.
• If you want, I can do a second-pass cleanup for old result subfolders (for example only keep recheck/ or only keep latest runs).