Rethinking

2024-11-09【人脸伪造检测】Spatial-Phase Shallow Learning: Rethinking Face Forgery Detection in Frequency Domain
一、研究动机[!note]创新点：利用相位谱实现伪造检测，并且证明了卷积模型可以提取隐性特征。由于上采样是伪造模型的关键步骤，这篇论文通过相位信息检测上采样的伪影。对比之前的频率模型：F3-Net：通过离散余弦变换后的统计特征实现伪造检测二、检测模型可学习的知识点
2024-10-29Rethinking Network Design and Local Geometry in Point Cloud:A Simple Residual MLP Framework——点云论文阅读9
此内容是论文总结，重点看思路！！文章概述本文提出了一种用于点云分析的简单残差MLP网络（PointMLP），通过省略复杂的几何特征提取器，仅采用残差MLP和轻量化的几何仿射模块，便能高效地提取点云特征，实现优异的分类性能。PointMLP在推理速度和准确性上优于许多现有方法，提供了一种更加高效的
2024-10-12PyTorchStepByStep - Chapter 2: Rethinking the Training Loop
defmake_train_step_fn(model,loss_fn,optimizer):defperform_train_step_fn(x,y):#SetmodeltoTRAINmodemodel.train()#Step1-Computemodel'spredictions-forwardpassyhat=model(x)
2023-12-18SegNeXt: Rethinking Convolutional Attention Design for Semantic Segmentation
SegNeXt:RethinkingConvolutionalAttentionDesignforSemanticSegmentation*Authors:[[Meng-HaoGuo]],[[Cheng-ZeLu]],[[QibinHou]],[[ZhengningLiu]],[[Ming-MingCheng]],[[Shi-MinHu]]·······初读印象comment::发现了导致分割模型性能提高的几
2023-12-18Rethinking and Improving Relative Position Encoding for Vision Transformer: ViT中的位置编码
RethinkingandImprovingRelativePositionEncodingforVisionTransformer*Authors:[[KanWu]],[[HouwenPeng]],[[MinghaoChen]],[[JianlongFu]],[[HongyangChao]]初读印象comment::(iRPE)提出了专门用于图像的相对位置编码方法，code:Cream/iRPEatmain·mi
2023-10-08Rethinking Point Cloud Registration as Masking and Reconstruction论文阅读
RethinkingPointCloudRegistrationasMaskingandReconstruction2023ICCV*GuangyanChen,MeilingWang,LiYuan,YiYang,YufengYue*;ProceedingsoftheIEEE/CVFInternationalConferenceonComputerVision(ICCV),2023,pp.17717-17727paper:Rethin
2023-07-20Rethinking with Retrieval Faithful Large Language Model Inference
目录概Rethinkingwithretrieval(RR)代码HeH.,ZhangH.andRothD.Rethinkingwithretrieval:faithfullargelanguagemodelinference.arXivpreprintarXiv:2301.00303,2023.概LLM(LargeLanguageModel)+检索.Rethinkingwithretrieval(RR)CoT(Chai
2023-03-03Rethinking the Heatmap Regression for Bottom-up Human Pose Estimation
本文的主要思想就是对heatmap图进行一个权重缩放。weight是作者提出的一个思路让模型将低输出值的位置加大权重，高输出值给予小权重，低输出值给与大全中。scaled_gt就是scale
2023-02-27Rethinking CNN Models for Audio Classification
WhatenablestheImageNetpretrainedmodelstolearnusefulaudiorepresentations,wesystematicallystudyhowmuchofpretrainedweightsisusefulforlearnin
2022-12-11论文推荐：Rethinking Attention with Performers
重新思考的注意力机制，Performers是由谷歌，剑桥大学，DeepMind，和艾伦图灵研究所发布在2021ICLR的论文已经超过500次引用传统的Transformer的使用softmax注意力，具有二次空间
2022-12-03【NeurIPS2022】ScalableViT: Rethinking the Context-oriented Generalization of Vision Transformer
【NeurIPS2022】ScalableViT:RethinkingtheContext-orientedGeneralizationofVisionTransformer这篇论文来自清华大学深圳研究生院和字节跳动。从Swin开始，attenti
2022-11-08【神经网络架构】EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks 论文阅读
原始题目EfficientNet:RethinkingModelScalingforConvolutionalNeuralNetworks中文名称EfficientNet:反思用于CNNs的模型扩展发表时间2019年5月28
2022-10-25[论文阅读] Rethinking the Truly Unsupervised Image-to-Image Translation
pretitle:RethinkingtheTrulyUnsupervisedImage-to-ImageTranslationaccepted:ICCV2021paper:arxiv|ICCVcode:https://github.com/clovaai/tunitref:htt
2022-10-05【NeurIPS 2022】SegNeXt: Rethinking Convolutional Attention Design for Semantic Segmentation
【NeurIPS2022】SegNeXt:RethinkingConvolutionalAttentionDesignforSemanticSegmentation代码：https://github.com/Visual-Attention-Network/SegNeXt1、研究