Anonymous Review Dataset This repository contains an evaluation dataset package for anonymous peer review. Overview This dataset contains 32,541 evaluation instances across 4 tasks , covering diverse visual domains (animal, plant, food, fungus, artifact, transport, environment, etc.). Task File Instances Metric MCQ mcq.jsonl 14,717 Accuracy Grid Sel (2x2) sel 2x2.jsonl 4,085 Accuracy Grid Sel (3x3) sel 3x3.jsonl 4,125 Accuracy Grid Loc (2x2) loc 2x2.jsonl 3,959 IoU@0.5 Grid Loc (3x3) loc 3x3.jsonl 3,446 IoU@0.5 Open QA open qa.jsonl 2,209 EM + LLM Judge Overall Score = mean(MCQ, Sel Avg, Loc Avg, Open QA corrected) Sel Avg = (Grid Sel 2x2 + Grid Sel 3x3) / 2 Loc Avg = (Grid Loc 2x2 + Grid Loc 3x3) / 2 Directory Structure Data Format MCQ ( mcq.jsonl ) Multiple choice question with 20 options. options : list of 20 candidate names gt index : index of the correct answer in options Grid Sel ( sel 2x2.jsonl , sel 3x3.jsonl ) Grid cell selection: identify which cell contains the target object. Grid Loc ( loc 2x2.jsonl , loc 3x3.jsonl ) Bounding box prediction: locate the target object with a bounding box. bbox norm : ground truth bounding box [x1, y1, x2, y2] normalized to [0, 1000] (0, 0…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy