code/postereval/configs/semantic_ir.example.json

{
  "run_name": "semantic_ir_evaluation",
  "method_order": ["method_a", "method_b"],
  "metrics": ["order", "completeness", "lta", "claim_f1"],
  "common_only": true,
  "prefix_alias_min_chars": null,
  "claim_model": "qwen3-vl-235b",
  "claim_match_mode": "strict_v2_t05_subset_numeric_one_side85",
  "lta": {
    "model_path": "<PATH_TO_QWEN3_VL_EMBEDDING_MODEL>"
  },
  "gt": {
    "content_ir_root": "<PATH_TO_GROUND_TRUTH_CONTENT_IR_ROOT>"
  },
  "methods": {
    "method_a": {
      "content_ir_root": "<PATH_TO_METHOD_A_CONTENT_IR_ROOT>",
      "figure_ir_root": "<PATH_TO_METHOD_A_FIGURE_IR_ROOT>",
      "poster_image_root": "<PATH_TO_METHOD_A_POSTER_IMAGE_ROOT>",
      "image_filename": "poster.png"
    },
    "method_b": {
      "content_ir_root": "<PATH_TO_METHOD_B_CONTENT_IR_ROOT>",
      "figure_ir_root": "<PATH_TO_METHOD_B_FIGURE_IR_ROOT>",
      "poster_image_root": "<PATH_TO_METHOD_B_POSTER_IMAGE_ROOT>",
      "image_filename": "poster.png"
    }
  }
}