{
  "image": {
    "path": "data/synthetic_ticket.ppm",
    "height": 8,
    "width": 8,
    "channels": 3
  },
  "patch_size": 2,
  "patch_rows": 4,
  "patch_cols": 4,
  "visual_token_count": 16,
  "attention_pairs": 256,
  "projection_dimension": 6,
  "patches": [
    {
      "patch_id": 0,
      "row": 0,
      "col": 0,
      "x": 0,
      "y": 0,
      "size": 2,
      "mean_rgb": [
        72.5,
        72.5,
        72.5
      ],
      "embedding_preview": [
        -0.5,
        -0.4463,
        -0.3925,
        -0.5969,
        -0.5431,
        -0.4894
      ]
    },
    {
      "patch_id": 1,
      "row": 0,
      "col": 1,
      "x": 2,
      "y": 0,
      "size": 2,
      "mean_rgb": [
        125.0,
        125.0,
        125.0
      ],
      "embedding_preview": [
        -0.44,
        -0.3239,
        -0.2078,
        -0.638,
        -0.522,
        -0.4059
      ]
    },
    {
      "patch_id": 2,
      "row": 0,
      "col": 2,
      "x": 4,
      "y": 0,
      "size": 2,
      "mean_rgb": [
        72.5,
        72.5,
        72.5
      ],
      "embedding_preview": [
        -0.38,
        -0.3663,
        -0.3525,
        -0.5969,
        -0.5831,
        -0.5694
      ]
    },
    {
      "patch_id": 3,
      "row": 0,
      "col": 3,
      "x": 6,
      "y": 0,
      "size": 2,
      "mean_rgb": [
        20.0,
        20.0,
        20.0
      ],
      "embedding_preview": [
        -0.32,
        -0.4086,
        -0.4973,
        -0.5557,
        -0.6443,
        -0.7329
      ]
    },
    {
      "patch_id": 4,
      "row": 1,
      "col": 0,
      "x": 0,
      "y": 2,
      "size": 2,
      "mean_rgb": [
        236.25,
        236.25,
        236.25
      ],
      "embedding_preview": [
        -0.46,
        -0.1894,
        0.0812,
        -0.6653,
        -0.3947,
        -0.1241
      ]
    },
    {
      "patch_id": 5,
      "row": 1,
      "col": 1,
      "x": 2,
      "y": 2,
      "size": 2,
      "mean_rgb": [
        227.5,
        227.5,
        227.5
      ],
      "embedding_preview": [
        -0.4,
        -0.1631,
        0.0737,
        -0.6584,
        -0.4216,
        -0.1847
      ]
    },
    {
      "patch_id": 6,
      "row": 1,
      "col": 2,
      "x": 4,
      "y": 2,
      "size": 2,
      "mean_rgb": [
        236.25,
        236.25,
        236.25
      ],
      "embedding_preview": [
        -0.34,
        -0.1094,
        0.1212,
        -0.6653,
        -0.4347,
        -0.2041
      ]
    },
    {
      "patch_id": 7,
      "row": 1,
      "col": 3,
      "x": 6,
      "y": 2,
      "size": 2,
      "mean_rgb": [
        245.0,
        245.0,
        245.0
      ],
      "embedding_preview": [
        -0.28,
        -0.0557,
        0.1686,
        -0.6722,
        -0.4478,
        -0.2235
      ]
    },
    {
      "patch_id": 8,
      "row": 2,
      "col": 0,
      "x": 0,
      "y": 4,
      "size": 2,
      "mean_rgb": [
        227.5,
        227.5,
        227.5
      ],
      "embedding_preview": [
        -0.42,
        -0.2031,
        0.0137,
        -0.5984,
        -0.3816,
        -0.1647
      ]
    },
    {
      "patch_id": 9,
      "row": 2,
      "col": 1,
      "x": 2,
      "y": 4,
      "size": 2,
      "mean_rgb": [
        255.0,
        255.0,
        255.0
      ],
      "embedding_preview": [
        -0.36,
        -0.12,
        0.12,
        -0.62,
        -0.38,
        -0.14
      ]
    },
    {
      "patch_id": 10,
      "row": 2,
      "col": 2,
      "x": 4,
      "y": 4,
      "size": 2,
      "mean_rgb": [
        227.5,
        227.5,
        227.5
      ],
      "embedding_preview": [
        -0.3,
        -0.1231,
        0.0537,
        -0.5984,
        -0.4216,
        -0.2447
      ]
    },
    {
      "patch_id": 11,
      "row": 2,
      "col": 3,
      "x": 6,
      "y": 4,
      "size": 2,
      "mean_rgb": [
        245.0,
        245.0,
        245.0
      ],
      "embedding_preview": [
        -0.24,
        -0.0557,
        0.1286,
        -0.6122,
        -0.4278,
        -0.2435
      ]
    },
    {
      "patch_id": 12,
      "row": 3,
      "col": 0,
      "x": 0,
      "y": 6,
      "size": 2,
      "mean_rgb": [
        245.0,
        245.0,
        245.0
      ],
      "embedding_preview": [
        -0.38,
        -0.1757,
        0.0286,
        -0.5522,
        -0.3478,
        -0.1435
      ]
    },
    {
      "patch_id": 13,
      "row": 3,
      "col": 1,
      "x": 2,
      "y": 6,
      "size": 2,
      "mean_rgb": [
        245.0,
        245.0,
        238.75
      ],
      "embedding_preview": [
        -0.3249,
        -0.1259,
        0.0388,
        -0.5473,
        -0.3825,
        -0.1835
      ]
    },
    {
      "patch_id": 14,
      "row": 3,
      "col": 2,
      "x": 4,
      "y": 6,
      "size": 2,
      "mean_rgb": [
        137.5,
        137.5,
        232.5
      ],
      "embedding_preview": [
        -0.1855,
        -0.4133,
        -0.1196,
        -0.5424,
        -0.2486,
        -0.4765
      ]
    },
    {
      "patch_id": 15,
      "row": 3,
      "col": 3,
      "x": 6,
      "y": 6,
      "size": 2,
      "mean_rgb": [
        191.25,
        191.25,
        245.0
      ],
      "embedding_preview": [
        -0.1578,
        -0.2243,
        0.0043,
        -0.5522,
        -0.3235,
        -0.39
      ]
    }
  ],
  "normalized_preview": [
    {
      "patch_id": 0,
      "normalized_mean_rgb": [
        -0.8764,
        -0.7665,
        -0.5408
      ]
    },
    {
      "patch_id": 1,
      "normalized_mean_rgb": [
        0.0227,
        0.1527,
        0.3742
      ]
    },
    {
      "patch_id": 2,
      "normalized_mean_rgb": [
        -0.8764,
        -0.7665,
        -0.5408
      ]
    },
    {
      "patch_id": 3,
      "normalized_mean_rgb": [
        -1.7754,
        -1.6856,
        -1.4559
      ]
    },
    {
      "patch_id": 4,
      "normalized_mean_rgb": [
        1.9278,
        2.1003,
        2.3132
      ]
    },
    {
      "patch_id": 5,
      "normalized_mean_rgb": [
        1.778,
        1.9471,
        2.1607
      ]
    }
  ],
  "resolution_budgets": [
    {
      "name": "mini_demo",
      "why": "ver patches a simple vista",
      "height": 8,
      "width": 8,
      "patch_size": 2,
      "patch_rows": 4,
      "patch_cols": 4,
      "visual_tokens": 16,
      "attention_pairs": 256,
      "padded_height": 8,
      "padded_width": 8,
      "padding_ratio": 0.0,
      "token_ratio_vs_vit_224_p16": 0.0816,
      "attention_pair_ratio_vs_vit_224_p16": 0.0067
    },
    {
      "name": "vit_base_224_p16",
      "why": "configuracion clasica de ViT base",
      "height": 224,
      "width": 224,
      "patch_size": 16,
      "patch_rows": 14,
      "patch_cols": 14,
      "visual_tokens": 196,
      "attention_pairs": 38416,
      "padded_height": 224,
      "padded_width": 224,
      "padding_ratio": 0.0,
      "token_ratio_vs_vit_224_p16": 1.0,
      "attention_pair_ratio_vs_vit_224_p16": 1.0
    },
    {
      "name": "captura_hd_p16",
      "why": "captura de pantalla relativamente comun",
      "height": 720,
      "width": 1280,
      "patch_size": 16,
      "patch_rows": 45,
      "patch_cols": 80,
      "visual_tokens": 3600,
      "attention_pairs": 12960000,
      "padded_height": 720,
      "padded_width": 1280,
      "padding_ratio": 0.0,
      "token_ratio_vs_vit_224_p16": 18.3673,
      "attention_pair_ratio_vs_vit_224_p16": 337.3594
    },
    {
      "name": "captura_hd_p32",
      "why": "menos tokens, mas compresion espacial",
      "height": 720,
      "width": 1280,
      "patch_size": 32,
      "patch_rows": 23,
      "patch_cols": 40,
      "visual_tokens": 920,
      "attention_pairs": 846400,
      "padded_height": 736,
      "padded_width": 1280,
      "padding_ratio": 0.022222,
      "token_ratio_vs_vit_224_p16": 4.6939,
      "attention_pair_ratio_vs_vit_224_p16": 22.0325
    },
    {
      "name": "captura_beca_larga_p16",
      "why": "captura larga de formulario completo: mucha evidencia pero coste alto",
      "height": 1440,
      "width": 3200,
      "patch_size": 16,
      "patch_rows": 90,
      "patch_cols": 200,
      "visual_tokens": 18000,
      "attention_pairs": 324000000,
      "padded_height": 1440,
      "padded_width": 3200,
      "padding_ratio": 0.0,
      "token_ratio_vs_vit_224_p16": 91.8367,
      "attention_pair_ratio_vs_vit_224_p16": 8433.9858
    },
    {
      "name": "captura_beca_region_alerta_p16",
      "why": "recorte de region con alerta y boton: menos tokens si sabes que evidencia necesitas",
      "height": 320,
      "width": 512,
      "patch_size": 16,
      "patch_rows": 20,
      "patch_cols": 32,
      "visual_tokens": 640,
      "attention_pairs": 409600,
      "padded_height": 320,
      "padded_width": 512,
      "padding_ratio": 0.0,
      "token_ratio_vs_vit_224_p16": 3.2653,
      "attention_pair_ratio_vs_vit_224_p16": 10.6622
    },
    {
      "name": "captura_movil_larga_p16",
      "why": "pantalla movil larga: aspect ratio extremo y mucho padding potencial",
      "height": 2532,
      "width": 1170,
      "patch_size": 16,
      "patch_rows": 159,
      "patch_cols": 74,
      "visual_tokens": 11766,
      "attention_pairs": 138438756,
      "padded_height": 2544,
      "padded_width": 1184,
      "padding_ratio": 0.016762,
      "token_ratio_vs_vit_224_p16": 60.0306,
      "attention_pair_ratio_vs_vit_224_p16": 3603.6744
    },
    {
      "name": "documento_alta_res_p16",
      "why": "pagina A4 escaneada a resolucion alta",
      "height": 1654,
      "width": 2339,
      "patch_size": 16,
      "patch_rows": 104,
      "patch_cols": 147,
      "visual_tokens": 15288,
      "attention_pairs": 233722944,
      "padded_height": 1664,
      "padded_width": 2352,
      "padding_ratio": 0.011637,
      "token_ratio_vs_vit_224_p16": 78.0,
      "attention_pair_ratio_vs_vit_224_p16": 6084.0
    },
    {
      "name": "factura_region_total_p16",
      "why": "recorte de tabla o total: alternativa a mandar toda la pagina",
      "height": 320,
      "width": 640,
      "patch_size": 16,
      "patch_rows": 20,
      "patch_cols": 40,
      "visual_tokens": 800,
      "attention_pairs": 640000,
      "padded_height": 320,
      "padded_width": 640,
      "padding_ratio": 0.0,
      "token_ratio_vs_vit_224_p16": 4.0816,
      "attention_pair_ratio_vs_vit_224_p16": 16.6597
    }
  ],
  "engineering_decision": {
    "rule": "bajar patch_size aumenta detalle, pero tambien tokens visuales y coste cuadratico de atencion",
    "watch": [
      "visual_tokens",
      "attention_pairs",
      "padding_ratio",
      "texto_pequeno",
      "aspect_ratio"
    ]
  },
  "valid": true,
  "issues": []
}