PosterText-30K / code /postereval /MANIFEST.md
HGTasd's picture
Initial clean release
ec21fa4 verified
# Release Manifest
This is the public evaluation-code entry for PosterEval-style poster metrics.
## Included
- `evaluate_structural_pptx.py`: PPTX evaluator for `Ove`, `Ali`, and `Ofl`.
- `prepare_pptx_autofit.py`: optional helper for materializing PPTX text-frame
autofit geometry before structural evaluation.
- `prepare_ir.py`: VLM-based IR generation from poster images.
- `evaluate_semantic_ir.py`: evaluator for `Order`, `Completeness`, `LTA`, and
`Claim F1`.
- `openrouter_client.py`: OpenAI-compatible OpenRouter JSON client.
- `qwen3_vl_embedding.py`: lightweight local wrapper used by `LTA`.
- `download_qwen3_vl_embedding.py`: helper for downloading
`Qwen/Qwen3-VL-Embedding-2B` into `models/modelscope/`.
- `prompts/`: public prompts for content IR, figure IR, and claim-pair scoring.
- `configs/semantic_ir.example.json`: generic placeholder config for content
metrics.
- `requirements.txt`: base Python dependency list.
## Excluded
- Raw paper PDFs, poster PPTX files, rendered poster images, and datasets.
- API keys, private environment files, local workspace paths, and runtime
artifacts.
- Paper-specific result tables or internal run outputs.