Skip to content

README_figure_mapping.md

Latest commit

 

History

History
54 lines (43 loc) · 3.37 KB

README_figure_mapping.md

File metadata and controls

54 lines (43 loc) · 3.37 KB
Raw

CaptchaBench — Figure Script Mapping

All Python scripts here generate the visualization figures. Script names match the figure numbers they produce.

Main Figures

Script Output PDF Section Description
figmm_fig1_teaser.py figures/fig1_teaser.pdf §1 Introduction Teaser: 5-VLM comparison with Gemini TVR anomaly highlight
figmm_fig2_pipeline_pptx.py figures/fig2_pipeline.pdf §3.2 Benchmark Dataset construction pipeline (ID ControlNet + SDXL)
figmm_fig3_radar.py figures/fig3_modality_radar.pdf §4.1 Baseline Eval Per-VLM radar: CRR/1-TVR/ASR for Qwen-VL, Gemini-3.0, GPT-5.2
figmm_fig4_pareto.py figures/fig4_pareto_efficiency.pdf §4.1 Baseline Eval Quality–Effectiveness Pareto Frontier (ID + SDXL, 6 methods)

Appendix Figures

Script Output PDF(s) Section Description
figmm_fig3_radar.py figures/fig7_radar_appendix.pdf App S6 Full 5-VLM radar appendix (Kimi 2.5 + GLM-4V extension)
figmm_fig6a_stroke_line.py figures/fig6a_stroke_line.pdf App S5 Stroke complexity line chart (ASR vs stroke-count bin)
figmm_fig6b_stroke_heatmap.py figures/fig6b_stroke_heatmap.pdf App S5 Stroke complexity heatmap (method × stroke-bin × VLM)
figmm_fig7_radar.py figures/fig7_radar_appendix.pdf App S6 Alternative radar appendix (older version)
figmm_fig8_vlm_bar_v2.py figures/fig8_vlm_bar_ID.pdf + fig8_vlm_bar_SDXL.pdf App S3 Per-VLM ASR grouped bar charts (ID and SDXL separately)
figmm_fig9a_case_kui.py figures/fig9a_case_kui.pdf App S8 Case study: 揆 character (22 strokes, enclosure type)
figmm_fig9b_case_jian.py figures/fig9b_case_jian.pdf App S8 Case study: 剑 character (9 strokes, left-right type)
figmm_fig10_case_study_simple.py figures/fig10_case_study_simple.pdf App S8 Case study: simple characters (6+11 strokes) — all protected
figmm_fig11_case_study_stroke.py figures/fig11_case_study_stroke.pdf App S8 Stroke-complexity case study across all 5 bins

Analysis Scripts (no figure output)

Script Purpose
compute_table4mm.py Compute modality-group analysis tables (Tab S2)
_find_diverse_case.py Helper: find diverse case study examples from eval results

Unused/Legacy Scripts

Script Status Notes
figmm_fig5_vlm_bar.py Legacy Single-panel version, replaced by figmm_fig8_vlm_bar_v2.py
figmm_fig5_quality.py Legacy Quality assessment figure (no longer in main figures)
figmm_fig5_quality_split.py Legacy Quality split version
figmm_fig6_stroke.py Legacy Combined stroke figure, replaced by fig6a + fig6b
figmm_fig7_radar.py Legacy Older radar version, superseded by figmm_fig3_radar.py
fig2_pipeline_v1.py Legacy First version of pipeline figure
figmm_case_study_grid.py Legacy Old case study grid layout
figmm_case_study.py Legacy Old case study script
figmm1_modality_radar.py Legacy Old modality radar version

Note on figmm_fig3_radar.py

This single script generates TWO figures:

  1. fig3_modality_radar.pdf — 3-VLM version for main text (Qwen-VL, Gemini-3.0, GPT-5.2)
  2. fig7_radar_appendix.pdf — full 5-VLM version for appendix