传统的Few-Shot Fine-Grained Image Classification (FS-FGIC)方法主要关注如何更好地提取细粒度特征,而忽视了在这个过程中可能引入的琐碎和不可泛化的样本级别噪声。这些噪声会导致模型在Few-Shot Learning (FSL)设置下的过拟合问题。文章提出了需要一种方法来提取具有辨别力的特征,同时抑制样本级别的噪声,以克服过拟...
1. generic object recognition -- mini-ImageNet 2. fine-grained image classification -- CUB 3. cross-domain adaptation -- mini-ImageNet -> CUB 结果翻译过来就是--换成cosine similarity classifier的baseline就非常的有用,比你们的有些还有用。 然后,我们现在才用了4-layer backbone(Conv-4 backbone),...
BSNet: Bi-Similarity Network for Few-shot Fine-grained Image Classification笔记,程序员大本营,技术文章内容聚合第一站。
Revisiting Local Descriptor based Image-to-Class Measure for Few-shot Learning 基于局部描述子的图像到类度量的少图像学习 Few-Shot Learning with Localization in Realistic Settings 真实场景中的少图像定位学习 Dense Classification and Implanting for Few-Shot Learning 密集分类与植入少镜头学习 Variational Protot...
Fine-grained visual categorization using meta-learning optimization with sample selection of auxiliary ...
Meta-Reinforced Synthetic Data for One-Shot Fine-Grained Visual Recognition 方向:细粒度视觉分类 问题:使用现成的图像生成器来合成额外的训练图像,往往无助于提高少样本细粒度识别的准确率。 方法:生成网络合成图像数据,利用元学习的方法将其于真实数据混合,使用元图像增强网络模型训练。
Fine-grained visual classificationFew-shot fine-grained visual classification aims to identify fine-grained concepts with very few samples, which is widely used in many fields, such as the classification of different species of birds in biological research, and the identification of car models in ...
Task Discrepancy Maximization for Fine-grained Few-Shot Classification SuBeen Lee, WonJun Moon, Jae-Pil Heo* Sungkyunkwan University {leesb7426, wjun0830, jaepilheo}@skku.edu Abstract Recognizing discriminative details such as eyes and beaks is important for distinguishing ...
(一)基于Finetune 这种方法已被广泛地应用。获得一定量的标注数据,然后基于一个基础网络进行微调。 这个基础网络是通过含有丰富标签的大规模数据集获得的,比如imagenet,我们的淘宝电商数据,称为通用数据域。然后在特定数据域上进行训练。训练时,会固定基础网络部分的参数,对领域特定的网络参数进行训练(这里有很多训练的...
[9] Guy Bukchin, Eli Schwartz, Kate Saenko, Ori Shahar, Rogerio Feris, Raja Giryes, and Leonid Karlinsky. Fine-grained angular contrastive learning with coarse labels. In CVPR, 2021. [10] Yuanhong Xu, Qi Qian, Hao Li, Rong Jin, and Juhua Hu. Weakly supervised representation learning with...