US 12,730,829 B2
Recording medium, data gathering apparatus, and method for gathering data
Kanata Suzuki, Kawasaki (JP); Taro Sunagawa, Kawasaki (JP); Tomotake Sasaki, Kawasaki (JP); and Takashi Katoh, Kawasaki (JP)
Assigned to Fujitsu Limited, Kawasaki (JP)
Filed by Fujitsu Limited, Kawasaki (JP)
Filed on Mar. 20, 2024, as Appl. No. 18/610,415.
Application 18/610,415 is a continuation of application No. PCT/JP2021/035083, filed on Sep. 24, 2021.
Prior Publication US 2024/0232231 A1, Jul. 11, 2024
Int. Cl. G06F 16/28 (2019.01); G06F 18/214 (2023.01); G06N 20/00 (2019.01)
CPC G06F 16/285 (2019.01) [G06F 18/214 (2023.01); G06N 20/00 (2019.01)] 9 Claims
OG exemplary drawing
 
1. A non-transitory computer-readable recording medium storing a gathering program that causes one or more computers to execute a process, the process comprising:
performing, by a trained feature extraction model, data augmentation on unlabeled data being read from a storing device;
providing a first label to a group of augmented data generated by the data augmentation, the first label indicating that labels of the augmented data all match;
providing, when a label for one of data of the augmented data is determined, the label to one or more data each provided with the first label that is same as the first label of the one of data; and
controlling, based on a control parameter predicted by a machine learning model, a position and a posture of a data obtainer that obtains training data such that a possibility that the label or the first label is successfully provided becomes highest,
wherein the controlling comprises:
obtaining, by the data obtainer, an initial image in an initial posture;
preparing a plurality of candidate postures each represented by a posture parameter;
for each candidate posture, inputting the initial image and the posture parameter of the candidate posture into the machine learning model to output a predicted confidence level indicating whether label information is obtainable in the candidate posture;
identifying, based on the predicted confidence levels for the plurality of candidate postures, a candidate posture having a highest predicted confidence level; and
controlling the data obtainer to move from the initial posture toward the identified candidate posture while continuously obtaining a plurality of images as the training data.