Data Selection for training Semantic Segmentation CNNs with cross-dataset weak supervision
This addresses the data efficiency challenge for training semantic segmentation models in automated driving, but it is incremental as it builds on existing weak supervision techniques.
The paper tackles the problem of reducing the amount of weakly labeled data needed to train semantic segmentation CNNs by proposing two data selection methods, resulting in up to 100 times reduction for Open Images and up to 20 times for Cityscapes.
Training convolutional networks for semantic segmentation with strong (per-pixel) and weak (per-bounding-box) supervision requires a large amount of weakly labeled data. We propose two methods for selecting the most relevant data with weak supervision. The first method is designed for finding visually similar images without the need of labels and is based on modeling image representations with a Gaussian Mixture Model (GMM). As a byproduct of GMM modeling, we present useful insights on characterizing the data generating distribution. The second method aims at finding images with high object diversity and requires only the bounding box labels. Both methods are developed in the context of automated driving and experimentation is conducted on Cityscapes and Open Images datasets. We demonstrate performance gains by reducing the amount of employed weakly labeled images up to 100 times for Open Images and up to 20 times for Cityscapes.