# How Low Can We Go? Pixel Annotation for Semantic Segmentation

arXiv: https://arxiv.org/abs/2201.10448 · HTML: https://mlsanity.com/paper/2201.10448/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 24 of its 44 references that are in the map.
- First posted on arXiv 25 Jan 2022.

## Builds on (most cited first, 24 of 24)

- [The Cityscapes Dataset for Semantic Urban Scene Understanding](https://mlsanity.com/paper/1604.01685/) (2016)
- [Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation](https://mlsanity.com/paper/1802.02611/) (2018)
- [Pyramid Scene Parsing Network](https://mlsanity.com/paper/1612.01105/) (2016)
- [Self-training with Noisy Student improves ImageNet classification](https://mlsanity.com/paper/1911.04252/) (2019)
- [Deep Bayesian Active Learning with Image Data](https://mlsanity.com/paper/1703.02910/) (2017)
- [ResNeSt: Split-Attention Networks](https://mlsanity.com/paper/2004.08955/) (2020)
- [Rethinking ImageNet Pre-training](https://mlsanity.com/paper/1811.08883/) (2018)
- [Rethinking Pre-training and Self-training](https://mlsanity.com/paper/2006.06882/) (2020)
- [ScribbleSup: Scribble-Supervised Convolutional Networks for Semantic Segmentation](https://mlsanity.com/paper/1604.05144/) (2016)
- [What's the Point: Semantic Segmentation with Point Supervision](https://mlsanity.com/paper/1506.02106/) (2015)
- [Zero-Shot Semantic Segmentation](https://mlsanity.com/paper/1906.00817/) (2019)
- [Deep Interactive Object Selection](https://mlsanity.com/paper/1603.04042/) (2016)
- [On Regularized Losses for Weakly-supervised CNN Segmentation](https://mlsanity.com/paper/1803.09569/) (2018)
- ["Double-DIP": Unsupervised Image Decomposition via Coupled Deep-Image-Priors](https://mlsanity.com/paper/1812.00467/) (2018)
- [Normalized Cut Loss for Weakly-supervised CNN Segmentation](https://mlsanity.com/paper/1804.01346/) (2018)
- [Learning random-walk label propagation for weakly-supervised semantic segmentation](https://mlsanity.com/paper/1802.00470/) (2018)
- [ViewAL: Active Learning with Viewpoint Entropy for Semantic Segmentation](https://mlsanity.com/paper/1911.11789/) (2019)
- [Reinforced active learning for image segmentation](https://mlsanity.com/paper/2002.06583/) (2020)
- [Not All Pixels Are Equal: Difficulty-aware Semantic Segmentation via Deep Layer Cascade](https://mlsanity.com/paper/1704.01344/) (2017)
- [Semantic Redundancies in Image-Classification Datasets: The 10% You Don't Need](https://mlsanity.com/paper/1901.11409/) (2019)
- [A Comprehensive Study on Colorectal Polyp Segmentation with ResUNet++, Conditional Random Field and Test-Time Augmentation](https://mlsanity.com/paper/2107.12435/) (2021)
- [Interactive Full Image Segmentation by Considering All Regions Jointly](https://mlsanity.com/paper/1812.01888/) (2018)
- [Are All Training Examples Created Equal? An Empirical Study](https://mlsanity.com/paper/1811.12569/) (2018)
- [Budget-aware Semi-Supervised Semantic and Instance Segmentation](https://mlsanity.com/paper/1905.05880/) (2019)

## Abstract (arXiv)

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.