מושגי ליבה
合成画像と実世界画像の間に存在する「現実のギャップ」を解決するために、新しい手法を提案し、アイ・トラッキングモデルのセグメンテーション性能を向上させる。
סטטיסטיקה
合成画像と実世界画像の間の平均距離:0.023(CGAN)、0.005(SRCGAN)
平均IoUスコア:RITnet - 0.94±0.00、DANN - 0.93±0.01
ציטוטים
"Despite the promise that synthetic data brings with it, a major issue emerges – due to the imperfections underlying computer simulation and 3D graphics models, a “reality gap” or mismatch exists between the synthetic data produced by a simulated environment (i.e. synthetic eye images) and the real world."
"Our results have positive implications for reducing the cost and burden associated with capturing and manually labeling large quantities of real human eye data, which in turn also promotes data privacy."