Institute for Theoretical Physics, University of Wroclaw, pl. Maxa Borna 9, 50-343, Wrocław, Poland.
NeuroSYS, Rybacka 7, 53-656, Wrocław, Poland.
Sci Rep. 2022 Jun 22;12(1):10583. doi: 10.1038/s41598-022-14879-3.
The statistical properties of the density map (DM) approach to counting microbiological objects on images are studied in detail. The DM is given by U[Formula: see text]-Net. Two statistical methods for deep neural networks are utilized: the bootstrap and the Monte Carlo (MC) dropout. The detailed analysis of the uncertainties for the DM predictions leads to a deeper understanding of the DM model's deficiencies. Based on our investigation, we propose a self-normalization module in the network. The improved network model, called Self-Normalized Density Map (SNDM), can correct its output density map by itself to accurately predict the total number of objects in the image. The SNDM architecture outperforms the original model. Moreover, both statistical frameworks-bootstrap and MC dropout-have consistent statistical results for SNDM, which were not observed in the original model. The SNDM efficiency is comparable with the detector-base models, such as Faster and Cascade R-CNN detectors.
详细研究了 U[公式:见正文]-Net 计数图像中微生物对象的密度图（DM）方法的统计特性。利用了两种用于深度神经网络的统计方法：自举法和蒙特卡罗（MC）随机失活法。对 DM 预测不确定性的详细分析加深了对 DM 模型缺陷的理解。基于我们的研究，我们在网络中提出了一种自归一化模块。改进后的网络模型称为自归一化密度图（SNDM），可以通过自身校正输出密度图，从而准确预测图像中的总对象数。SNDM 架构优于原始模型。此外，自举法和 MC 随机失活法这两种统计框架对 SNDM 的统计结果是一致的，而在原始模型中并没有观察到这种一致性。SNDM 的效率可与基于检测器的模型（如 Faster 和 Cascade R-CNN 检测器）相媲美。
