images/1604.04939v1/gt.json

[
  {
    "id": 0,
    "page": 32,
    "bounding_box": [
      181.9040069580078,
      90.86201477050781,
      430.09600830078125,
      211.010986328125
    ],
    "latex_content": "\\begin{table}[h]\n\\begin{small}\n\\begin{center}\n\\begin{tabular}{ l | l }\n%\\hline\n                               & Error \\\\ \\hline %\\hline\nMean Training Pose             & 6.16   \\\\ %\\hline\nLinear Regression              & 5.86   \\\\ %\\hline\nGP Regression                  & 4.27   \\\\ %\\hline\nNearest Neighbour (sil. space) & 4.88  \\\\ %\\hline\nNearest Neighbour with sequences (sil. space) & 4.04  \\\\ %\\hline\n\\textcolor{light-gray}{Nearest Neighbour (pose space)} & \\textcolor{light-gray}{2.08}   \\\\ %\\hline\n%Nearest Neighbour (pose space) & 2.08   \\\\ %\\hline %---\nShared GP-LVM                      & 5.13    \\\\ % \\hline\nMRD without Dynamics       & 4.67   \\\\ %\\hline\nMRD with Dynamics          & \\textbf{2.94}    \\\\ \\hline\n\\end{tabular}\n\\end{center}\n\\end{small}\n\\caption[MRD, human pose experiment: MSE and comparisons.]{\nThe mean of the Euclidean distances of the joint locations between the predicted and the true poses. The nearest neighbour in the pose space is not a fair comparison (since the test pose is supposed to be unseen), but is reported as it provides some insight about the lower bound on the error that can be achieved for this task.\n}\n\\label{tab:humanMotionTable}\n\\end{table}",
    "extracted_content": [
      [
        "",
        "Error"
      ],
      [
        "Mean Training Pose\nLinear Regression\nGP Regression\nNearest Neighbour (sil. space)\nNearest Neighbour with sequences (sil. space)\nNearest Neighbour (pose space)\nShared GP-LVM\nMRD without Dynamics\nMRD with Dynamics",
        "6.16\n5.86\n4.27\n4.88\n4.04\n2.08\n5.13\n4.67\n2.94"
      ]
    ],
    "similarity_score": 0.38911290322580644,
    "table_image": "images/1604.04939v1/table_0.png",
    "page_image": "pages/1604.04939v1/page_32.png"
  },
  {
    "id": 1,
    "page": 36,
    "bounding_box": [
      182.42150115966797,
      477.52899169921875,
      429.57850646972656,
      521.7630004882812
    ],
    "latex_content": "\\begin{table}[h]\n\\begin{center}\n  \\begin{tabular}{ c | c || c | c | c}\n    \\hline\n    Given         & Predicted &   MRD &    NN &  GP  \\\\ \\hline \\hline\n    audio         & video     & 0.498 & 0.485 & 0.494 \\\\ \\hline\n    audio, labels & video     & 0.434 & 0.485 & 0.472 \\\\ \\hline\n  \\end{tabular}\n\\end{center}\n  \\caption[RMSE for test predictions on ``AVletters'': MRD and comparisons.]{RMSE of predicting the video information of test data, given only the audio or the audio and subject id information.}\n  \\label{table:MRD_AVletters}\n\\end{table}",
    "extracted_content": [
      [
        "Given",
        "Predicted",
        "MRD",
        "NN",
        "GP"
      ],
      [
        "audio",
        "video",
        "0.498",
        "0.485",
        "0.494"
      ],
      [
        "audio, labels",
        "video",
        "0.434",
        "0.485",
        "0.472"
      ]
    ],
    "similarity_score": 0.4708860759493671,
    "table_image": "images/1604.04939v1/table_1.png",
    "page_image": "pages/1604.04939v1/page_36.png"
  }
]