Data Volume = Foundational to Algorithm Refinement + Artificial Intelligence (AI) Performance… Object Detection - Performance vs. Dataset Size Google Research & Carnegie Mellon, 2017 ) 95]40 . @[.5, P (mA 35 n o Precisi n 30 ctio ete D ect j 25 Ob 10MM 30MM 100MM 10 100 300MM Example Images in Training Dataset Source: Revisiting Unreasonable Effectiveness of Data in Deep Learning Era – Sun, Shrivastava, Singh, & Gupta, 2017 Note: Chart reflects object detection performance when initial checkpoints are pre-trained on different subsets of JFT-300M tagged image 196 dataset. X-axis is the data size in log-scale, y-axis is the detection performance in mAP@[.5,.95] on “COCO minival” testing set.
