qq_49230659的博客

这篇工作在CSS领域主要是利用蒸馏和当前类别比较像的先前类别的知识，加强旧的容易被遗忘的类别的记忆，也让新类别学习得更好。作者这篇工作的观点在于，和新类别相似的旧类别在学习新类别时更容易被遗忘，即旧类别会被认为是新类别。...

continual learning旨在用一个模型有效解决增量任务的学习，这篇工作可以看做是基于重演的对抗知识遗忘的方法，提出了一个 Gradient Coreset Replay (GCR)策略来重演被选择的缓存，其中本文选择一个Coreset（核心集），这个核心集尽力近似所有见过的数据的模型的梯度。...

Representation Compensation Networks for Continual Semantic SegmentationAbstract1. Introduction2. Related Work3. Method3.1. Preliminaries论文地址github code很久没有更新paper reading的内容了，最近发现Continual Semantic Segmentation这个领域有点意思，在这里记录一下。Abstract这篇工作研究Continual

Meta-Semi: A Meta-learning Approach for Semi-supervised LearningAbstract1 Introduction论文地址Abstract半监督学习近年来得到广泛关注，然而他们往往引入多个超参数，但一般半监督问题有标注的数据是稀缺的，没有足够的标注用来调整这些超参数。本文提出了一种基于meta-learning的半监督算法，只用调整一种超参数就可以在很多半监督学习场景中获得不错的表现。我们定义了一个meta optimization probl

Learning What to Learn for Video Object SegmentationAbstract3 Method3.1 Video Object Segmentation as Few-shot Learning3.2 Learning What to Learn3.3 Internal Learner3.4 Video Object Segmentation Architecture论文地址codeAbstract通过meta-learning的思想完成one-shot V

Learning Calibrated Medical Image Segmentation via Multi-rater Agreement Modeling摘要1. Introduction2. Related Work3. Methodology3.1. Motivation3.2. Overall Framework论文地址code视频讲解摘要本文是关于分割在医学场景中的应用。一般的分割任务有确定的真值，但是对于一些医学任务，我们往往需要结合多个专家的诊断结果作为真值，传统的方法是平均各

Contour Knowledge Transfer for Salient Object Detection摘要论文地址code摘要本文是一篇无监督显著性监测方法，C2S-Net。

Weakly-Supervised Salient Object Detection via Scribble Annotations摘要1. Introduction2. Related Work2.1. Learning Saliency from Weak Annotations2.2. Weakly-Supervised Semantic Segmentation2.3. Recovering Structure from Weak Labels2.4. Comparison with Existi

DeepUSPS: Deep Robust Unsupervised Saliency Prediction With Self-Supervision摘要1 Introduction2 Related work3 DeepUSPS: Deep Unsupervised saliency prediction via self-supervision3.1 Enforcing inter-images consistency with image-level loss3.2 Incremental pseu

Integral Object Mining via Online Attention Accumulation摘要论文地址摘要

Splitting Vs. Merging: Mining Object Regions with Discrepancy and Intersection Loss for Weakly Supervised Semantic Segmentation摘要1 Introduction2 Related Works3 Approach3.1 Revisiting CAM3.2 Splitting vs. Merging论文地址摘要本文关注于image-level的WSSS任务。本文通过训练一个regi

Self-Supervised Difference Detectionfor Weakly-Supervised Semantic Segmentation摘要1. Introductionfor Weakly-Supervised Semantic Segmentation)论文地址摘要本文通过移除噪音来提升mapping function的准确性。本文提出self-supervised difference detection模块，通过预测mapping前后的分割掩码来减少noise。1.

Joint Learning of Saliency Detection and Weakly Supervised Semantic Segmentation摘要1. Introduction3. The proposed approach3.1. Network overview3.2. Segmentation networks3.3. Saliency aggregation3.4. Jointly learning of saliency and segmentation论文地址摘要本文通过

Mining Cross-Image Semantics for Weakly Supervised Semantic Segmentation摘要1 Introduction2 Related Work3 Methodology3.1 Co-attention Classification Network摘要本文研究只在image-level监督下学习语义分割。目前流行的解决方法是使用分类器生成的目标localization maps作为监督，致力于让localization maps捕捉更完整的目标

Employing Multi-estimations for Weakly-Supervised Semantic Segmentation摘要摘要

From Semantic Categories to Fixations: A Novel Weakly-supervised Visual-auditory Saliency Detection Approach摘要摘要多亏了深度学习和大规模训练数据集的快速发展，视频显著性检测模型的表现显著增强。然而，基于音视频眼动点预测的深度学习仍具有挑战性。目前，只有少数视音频序列具有在真实视音频环境中记录的真实注视。 因此，在相同的视音频环境下重新收集真实的注视既没有效率也没有必要。为了解决这样的问题，本文提

Weakly Supervised Video Salient Object Detection摘要3. Our Method3.1. Overview3.2. Fixation guided scribble annotation摘要本文首次提出了基于重标注的“眼动点模糊标注”的弱监督视频显著性目标检测模型。本文提出了“Appearance-motion fusion module”和双向的LSTM；另外还设计了前背景similarity loss；另外还提出了一个weak annotation bo

Railroad is not a Train: Saliency as Pseudo-pixel Supervision for Weakly Supervised Semantic Segmentation摘要3. Proposed Method3.1. Motivation3.2. Explicit Pseudo-pixel Supervision3.3. Joint Training Procedure摘要现有的弱监督语义分割（WSSS）使用image-level的弱监督，但这样有很多限制：稀疏

Non-Salient Region Object Mining for Weakly Supervised Semantic Segmentation摘要摘要

Weakly-supervised Instance Segmentation via Class-agnostic Learning with Salient Images摘要3. Method3.1. Weakly-supervised Instance Segmentation via Multi-instance Learning3.2. Joint Training with Salient Images3.2.1 Salient Images3.2.2 Data Augmentation3.3.