Model-Assisted Labeling and Self-Training for Label Noise Reduction in the Detection of Stains on Images of Laundry

Abstract

In this work, the creation of a dataset labeled in a pixel-wise manner for the uncommon domain of stain detection on patterned laundry is described. The unique properties of images in this dataset—stains are small and sometimes occur in large amounts—led to the creation of noisy labels. Indeed, the training of a fully convolutional neural network for salient object detection with this dataset revealed that the model predicts stains missed by human labelers. Thus, the reduction in label noise by adding overlooked regions with the help of the model’s predictions is examined in two different experiments. In the model-assisted labeling experiment, a simulation is ran where a human selects correct regions from the predictions. In the self-training experiment, regions of high certainty are automatically selected from the predictions. Re-training the model with the revised labels shows that model-assisted labeling leads to an average improvement in performance by 8.52%. In contrast, with self-training, the performance increase is generally lower (2.58% on average) and a decrease is even possible since regions of high certainty are often false positives.

Toward a generalizable machine learning workflow for neurodegenerative disease staging with focus on neurofibrillary tangles

“…Labeling large datasets, such as those needed for this study, is very time consuming and exhaustive, and experts generally have neither the freedom nor desire to commit extensive effort to this process. Fine-tuning labels using a well-developed application, such as that which was developed and demonstrated in this work, substantially reduces annotation time and annotator fatigue without sacrificing performance [ 63 , 64 ]. Furthermore, this workflow is generalizable; it is likely just as applicable to this kind of problem in the context of cancer as it has been shown to be in the case of neuropathology.…”

PhytoOracle: Scalable, modular phenomics data processing pipelines

“…To train segmentation ML models, a random sample of individual plant point clouds were collected and labeled using a model-assisted labeling (MAL) approach (Model-assisted labeling (MAL); Huxohl and Kummert, 2021 ). The MAL script fit a plane to each point cloud and resulted in the labeling of two classes: plant and soil (see Code Availability Statement).…”