Domain Adaptive Knowledge Distillation for Driving Scene Semantic Segmentation

Kothandaraman, Divya; Nambiar, Athira; Mittal, Anurag

doi:10.1109/wacvw52041.2021.00019

Cited by 22 publications

(10 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…where q s i is the prediction map of the current model M t on the current input x t for the previous task t − 1; q t i is the prediction map of the previous model M t−1 on the current input x t for the previous task t − 1; ϕ is the KL-divergence (KLD) loss between these two softmax probability distribution maps, computed and summed over each previously learned task i, 0 < i < t; λ KLD is the regularization hyperparameter for KL-divergence. KLD here effectively distills domain knowledge from the teacher q t i to the student q s i , and can be seen as domain adaptive knowledge distillation [22]. Total loss for domain-invariant parameters W s is hence given as:…”

Section: Domain-invariant Parametersmentioning

confidence: 99%

Multi-Domain Incremental Learning for Semantic Segmentation

Garg

Saluja

Balasubramanian

et al. 2022

2022 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)

View full text Add to dashboard Cite

Recent efforts in multi-domain learning for semantic segmentation attempt to learn multiple geographical datasets in a universal, joint model. A simple fine-tuning experiment performed sequentially on three popular road scene segmentation datasets demonstrates that existing segmentation frameworks fail at incrementally learning on a series of visually disparate geographical domains. When learning a new domain, the model catastrophically forgets previously learned knowledge. In this work, we pose the problem of multi-domain incremental learning for semantic segmentation. Given a model trained on a particular geographical domain, the goal is to (i) incrementally learn a new geographical domain, (ii) while retaining performance on the old domain, (iii) given that the previous domain's dataset is not accessible. We propose a dynamic architecture that assigns universally shared, domain-invariant parameters to capture homogeneous semantic features present in all domains, while dedicated domain-specific parameters learn the statistics of each domain. Our novel optimization strategy helps achieve a good balance between retention of old knowledge (stability) and acquiring new knowledge (plasticity). We demonstrate the effectiveness of our proposed solution on domain incremental settings pertaining to realworld driving scenes from roads of Germany (Cityscapes), the United States (BDD100k), and India (IDD).

show abstract

Section: Domain-invariant Parametersmentioning

confidence: 99%

Multi-Domain Incremental Learning for Semantic Segmentation

Garg

Saluja

Balasubramanian

et al. 2022

2022 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)

View full text Add to dashboard Cite

show abstract

“…In addition to the centroid-aware methods, there are also many trivial solutions for domain adaptation, e.g., Knowledge Distillation [54], [60] and Mixing [59], [61]. These popular methods indeed boost the performance to a record high, yet the training process is prone to be time-consuming and requires much computational resources, making unsupervised domain adaptation impractical in industrialized scenarios.…”

Section: Discussionmentioning

confidence: 99%

“…For the fine-grained feature alignment, these methods focus on reducing the distance between the corresponding classes of two domains. In addition, there are also many effective solutions for domain adaptation, e.g., Knowledge Distillation [54], [60] and Mixing [59], [61]. The proposal of these methods leads to considerable progress on major open benchmark datasets.…”

Section: B Unsupervised Domain Adaptationmentioning

confidence: 99%

Confidence-and-Refinement Adaptation Model for Cross-Domain Semantic Segmentation

Zhang

Chen

Shen

et al. 2022

IEEE Trans. Intell. Transport. Syst.

View full text Add to dashboard Cite

With the rapid development of convolutional neural networks (CNNs), significant progress has been achieved in semantic segmentation. Despite the great success, such deep learning approaches require large scale real-world datasets with pixel-level annotations. However, considering that pixel-level labeling of semantics is extremely laborious, many researchers turn to utilize synthetic data with free annotations. But due to the clear domain gap, the segmentation model trained with the synthetic images tends to perform poorly on the real-world datasets. Unsupervised domain adaptation (UDA) for semantic segmentation recently gains an increasing research attention, which aims at alleviating the domain discrepancy. Existing methods in this scope either simply align features or the outputs across the source and target domains or have to deal with the complex image processing and post-processing problems. In this work, we propose a novel multi-level UDA model named Confidenceand-Refinement Adaptation Model (CRAM), which contains a confidence-aware entropy alignment (CEA) module and a style feature alignment (SFA) module. Through CEA, the adaptation is done locally via adversarial learning in the output space, making the segmentation model pay attention to the high-confident predictions. Furthermore, to enhance the model transfer in the shallow feature space, the SFA module is applied to minimize the appearance gap across domains. Experiments on two challenging UDA benchmarks "GTA5-to-Cityscapes" and "SYNTHIA-to-Cityscapes" demonstrate the effectiveness of CRAM. We achieve comparable performance with the existing state-of-the-art works with advantages in simplicity and convergence speed.

show abstract

“…Furthermore, taking advantage of the intrinsic spatial structure presented in urban scene images (that they focus on), they propose to partition the images into non-overlapping grids, and the domain alignment is performed on the pixel-level features from the same spatial region using GRL [50]. Finally, The Domain Adaptive Knowledge Distillation model [88] consists in a multi-level strategy to effectively distill knowledge at different levels -feature space and output spaceusing a combination of KL divergence and MSE losses.…”

Section: Entropy Minimization Of Target Predictions (Tem)mentioning

confidence: 99%

Unsupervised Domain Adaptation for Semantic Image Segmentation: a Comprehensive Survey

Volpi¹,

Chidlovskii²

2021

Preprint

View full text Add to dashboard Cite

Semantic segmentation plays a fundamental role in a broad variety of computer vision applications, providing key information for the global understanding of an image. Yet, the state-of-the-art models rely on large amount of annotated samples, which are more expensive to obtain than in tasks such as image classification. Since unlabelled data is instead significantly cheaper to obtain, it is not surprising that Unsupervised Domain Adaptation reached a broad success within the semantic segmentation community.This survey is an effort to summarize five years of this incredibly rapidly growing field, which embraces the importance of semantic segmentation itself and a critical need of adapting segmentation models to new environments. We present the most important semantic segmentation methods; we provide a comprehensive survey on domain adaptation techniques for semantic segmentation; we unveil newer trends such as multi-domain learning, domain generalization, test-time adaptation or source-free domain adaptation; we conclude this survey by describing datasets and benchmarks most widely used in semantic segmentation research. We hope that this survey will provide researchers across academia and industry with a comprehensive reference guide and will help them in fostering new research directions in the field.1 See as example the https://github.com/errolPereira/ Semi-Automatic-Image-Annotation-Tool 2 We refer to the dedicated page https://github.com/ heartexlabs/awesome-data-labeling for an up-to-date collection of annotation tools.

show abstract

Domain Adaptive Knowledge Distillation for Driving Scene Semantic Segmentation

Cited by 22 publications

References 23 publications

Multi-Domain Incremental Learning for Semantic Segmentation

Multi-Domain Incremental Learning for Semantic Segmentation

Confidence-and-Refinement Adaptation Model for Cross-Domain Semantic Segmentation

Unsupervised Domain Adaptation for Semantic Image Segmentation: a Comprehensive Survey

Contact Info

Product

Resources

About