How Low Can We Go? Pixel Annotation for Semantic Segmentation
This addresses the high annotation cost problem for researchers and practitioners in computer vision, offering a method to significantly reduce labeling effort, though it is incremental as it builds on Active Learning and pseudo-labeling techniques.
The study tackled the problem of minimizing pixel annotation for semantic segmentation by using an Oracle with Active Learning, finding that less than 0.1% of pixels need annotation to achieve over 98% accuracy per image, and extended this to dataset-level annotation to reduce costs while matching full annotation performance.
How many labeled pixels are needed to segment an image, without any prior knowledge? We conduct an experiment to answer this question. In our experiment, an Oracle is using Active Learning to train a network from scratch. The Oracle has access to the entire label map of the image, but the goal is to reveal as little pixel labels to the network as possible. We find that, on average, the Oracle needs to reveal (i.e., annotate) less than 0.1% of the pixels in order to train a network. The network can then label all pixels in the image at an accuracy of more than 98%. Based on this single-image-annotation experiment, we design an experiment to quickly annotate an entire data set. In the data set level experiment the Oracle trains a new network for each image from scratch. The network can then be used to create pseudo-labels, which are the network predicted labels of the unlabeled pixels, for the entire image. Only then, a data set level network is trained from scratch on all the pseudo-labeled images at once. We repeat both image level and data set level experiments on two, very different, real-world data sets, and find that it is possible to reach the performance of a fully annotated data set using a fraction of the annotation cost.