Bias-Aware Loss for Training Image and Speech Quality Prediction Models from Multiple Datasets
This addresses the issue of inconsistent ground truth ratings in quality prediction models for researchers and engineers, though it is incremental as it builds on existing loss function methods.
The paper tackled the problem of experiment-specific biases in training image and speech quality prediction models from multiple datasets, which confuse neural networks and reduce performance, by proposing a bias-aware loss function that estimates dataset biases during training, resulting in improved model efficiency as validated on synthetic and subjective datasets.
The ground truth used for training image, video, or speech quality prediction models is based on the Mean Opinion Scores (MOS) obtained from subjective experiments. Usually, it is necessary to conduct multiple experiments, mostly with different test participants, to obtain enough data to train quality models based on machine learning. Each of these experiments is subject to an experiment-specific bias, where the rating of the same file may be substantially different in two experiments (e.g. depending on the overall quality distribution). These different ratings for the same distortion levels confuse neural networks during training and lead to lower performance. To overcome this problem, we propose a bias-aware loss function that estimates each dataset's biases during training with a linear function and considers it while optimising the network weights. We prove the efficiency of the proposed method by training and validating quality prediction models on synthetic and subjective image and speech quality datasets.