# Decomposed Knowledge Distillation for Class-Incremental Semantic Segmentation

arXiv: https://arxiv.org/abs/2210.05941 · HTML: https://mlsanity.com/paper/2210.05941/

## Key facts

- Cited by 28 papers in the map (data through 14 Oct 2026).
- Placed in Detection and dense prediction distillation, one of 12 directions in Knowledge distillation.
- Builds on 21 of its 35 references that are in the map.
- First posted on arXiv 12 Oct 2022.

## Placed in

- Knowledge distillation / [Detection and dense prediction distillation](https://mlsanity.com/distill/detection-and-dense-prediction-distillation/)

## Builds on (most cited first, 21 of 21)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [ImageNet Large Scale Visual Recognition Challenge](https://mlsanity.com/paper/1409.0575/) (2014)
- [iCaRL: Incremental Classifier and Representation Learning](https://mlsanity.com/paper/1611.07725/) (2016)
- [Rethinking Atrous Convolution for Semantic Image Segmentation](https://mlsanity.com/paper/1706.05587/) (2017)
- [Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour](https://mlsanity.com/paper/1706.02677/) (2017)
- [Large Scale Incremental Learning](https://mlsanity.com/paper/1905.13260/) (2019)
- [End-to-End Incremental Learning](https://mlsanity.com/paper/1807.09536/) (2018)
- [Holistically-Nested Edge Detection](https://mlsanity.com/paper/1504.06375/) (2015)
- [Learning without Memorizing](https://mlsanity.com/paper/1811.08051/) (2018)
- [One-Shot Video Object Segmentation](https://mlsanity.com/paper/1611.05198/) (2016)
- [Deeply supervised salient object detection with short connections](https://mlsanity.com/paper/1611.04849/) (2016)
- [Modeling the Background for Incremental Learning in Semantic Segmentation](https://mlsanity.com/paper/2002.00718/) (2020)
- [PLOP: Learning without Forgetting for Continual Semantic Segmentation](https://mlsanity.com/paper/2011.11390/) (2020)
- [Incremental Learning Techniques for Semantic Segmentation](https://mlsanity.com/paper/1907.13372/) (2019)
- [SS-IL: Separated Softmax for Incremental Learning](https://mlsanity.com/paper/2003.13947/) (2020)
- [Continual Semantic Segmentation via Repulsion-Attraction of Sparse and Disentangled Latent Representations](https://mlsanity.com/paper/2103.06342/) (2021)
- [RECALL: Replay-based Continual Learning in Semantic Segmentation](https://mlsanity.com/paper/2108.03673/) (2021)
- [Representation Compensation Networks for Continual Semantic Segmentation](https://mlsanity.com/paper/2203.05402/) (2022)
- [SSUL: Semantic Segmentation with Unknown Label for Exemplar-based Class-Incremental Learning](https://mlsanity.com/paper/2106.11562/) (2021)
- [Essentials for Class Incremental Learning](https://mlsanity.com/paper/2102.09517/) (2021)

## Built on by (most cited first, 28 of 28)

- [Gradient Reweighting: Towards Imbalanced Class-Incremental Learning](https://mlsanity.com/paper/2402.18528/) (2024)
- [A Survey on Continual Semantic Segmentation: Theory, Challenge, Method and Application](https://mlsanity.com/paper/2310.14277/) (2023)
- [Distill n' Explain: explaining graph neural networks using simple surrogates](https://mlsanity.com/paper/2303.10139/) (2023)
- [Rethinking Query-based Transformer for Continual Image Segmentation](https://mlsanity.com/paper/2507.07831/) (2025)
- [Gradient-Semantic Compensation for Incremental Semantic Segmentation](https://mlsanity.com/paper/2307.10822/) (2023)
- [Strike a Balance in Continual Panoptic Segmentation](https://mlsanity.com/paper/2407.16354/) (2024)
- [DiffusePast: Diffusion-based Generative Replay for Class Incremental Semantic Segmentation](https://mlsanity.com/paper/2308.01127/) (2023)
- [Taxonomy-Aware Continual Semantic Segmentation in Hyperbolic Spaces for Open-World Perception](https://mlsanity.com/paper/2407.18145/) (2024)
- [Cs2K: Class-specific and Class-shared Knowledge Guidance for Incremental Semantic Segmentation](https://mlsanity.com/paper/2407.09047/) (2024)
- [Early Preparation Pays Off: New Classifier Pre-tuning for Class Incremental Semantic Segmentation](https://mlsanity.com/paper/2407.14142/) (2024)
- [Learning at a Glance: Towards Interpretable Data-limited Continual Semantic Segmentation via Semantic-Invariance Modelling](https://mlsanity.com/paper/2407.15429/) (2024)
- [Beyond Background Shift: Rethinking Instance Replay in Continual Semantic Segmentation](https://mlsanity.com/paper/2503.22136/) (2025)
- [CoMBO: Conflict Mitigation via Branched Optimization for Class Incremental Segmentation](https://mlsanity.com/paper/2504.04156/) (2025)
- [Background Adaptation with Residual Modeling for Exemplar-Free Class-Incremental Semantic Segmentation](https://mlsanity.com/paper/2407.09838/) (2024)
- [Continual Panoptic Perception: Towards Multi-modal Incremental Interpretation of Remote Sensing Images](https://mlsanity.com/paper/2407.14242/) (2024)
- [Dynamic Robot-Assisted Surgery with Hierarchical Class-Incremental Semantic Segmentation](https://mlsanity.com/paper/2508.01713/) (2025)
- [ProtoFlow: Mitigating Forgetting in Class-Incremental Remote Sensing Segmentation via Low-Curvature Prototype Flow](https://mlsanity.com/paper/2604.03212/) (2026)
- [BACS: Background Aware Continual Semantic Segmentation](https://mlsanity.com/paper/2404.13148/) (2024)
- [CFSSeg: Closed-Form Solution for Class-Incremental Semantic Segmentation of 2D Images and 3D Point Clouds](https://mlsanity.com/paper/2412.10834/) (2024)
- [Decoupling Continual Semantic Segmentation](https://mlsanity.com/paper/2508.05065/) (2025)
- [Towards Realistic Incremental Scenario in Class Incremental Semantic Segmentation](https://mlsanity.com/paper/2405.09858/) (2024)
- [Completely Weakly Supervised Class-Incremental Learning for Semantic Segmentation](https://mlsanity.com/paper/2505.10781/) (2025)
- [Learning Yourself: Class-Incremental Semantic Segmentation with Language-Inspired Bootstrapped Disentanglement](https://mlsanity.com/paper/2509.00527/) (2025)
- [Evolving Knowledge Mining for Class Incremental Segmentation](https://mlsanity.com/paper/2306.02027/) (2023)
- [Revisiting Continual Semantic Segmentation with Pre-trained Vision Models](https://mlsanity.com/paper/2508.04267/) (2025)
- [Evolving Without Ending: Unifying Multimodal Incremental Learning for Continual Panoptic Perception](https://mlsanity.com/paper/2601.15643/) (2026)
- [PILOT: A Data-Free Continual Learning Approach for Real-Time Semantic Segmentation via Boundary Guidance](https://mlsanity.com/paper/2605.27128/) (2026)
- [Learning Probabilistic Prompt for Continual Learning](https://mlsanity.com/paper/2607.04711/) (2026)

## Abstract (arXiv)

Class-incremental semantic segmentation (CISS) labels each pixel of an image with a corresponding object/stuff class continually. To this end, it is crucial to learn novel classes incrementally without forgetting previously learned knowledge. Current CISS methods typically use a knowledge distillation (KD) technique for preserving classifier logits, or freeze a feature extractor, to avoid the forgetting problem. The strong constraints, however, prevent learning discriminative features for novel classes. We introduce a CISS framework that alleviates the forgetting problem and facilitates learning novel classes effectively. We have found that a logit can be decomposed into two terms. They quantify how likely an input belongs to a particular class or not, providing a clue for a reasoning process of a model. The KD technique, in this context, preserves the sum of two terms (i.e., a class logit), suggesting that each could be changed and thus the KD does not imitate the reasoning process. To impose constraints on each term explicitly, we propose a new decomposed knowledge distillation (DKD) technique, improving the rigidity of a model and addressing the forgetting problem more effectively. We also introduce a novel initialization method to train new classifiers for novel classes. In CISS, the number of negative training samples for novel classes is not sufficient to discriminate old classes. To mitigate this, we propose to transfer knowledge of negatives to the classifiers successively using an auxiliary classifier, boosting the performance significantly. Experimental results on standard CISS benchmarks demonstrate the effectiveness of our framework.