Attribute-Based Classification for Zero-Shot Visual Object Categorization

By Christoph H. Lampert, Hannes Nickisch, Stefan Harmeling

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Unbounded cache model for online language modeling with open vocabulary (2017, background): “Zero-shot learning [41] can deal with new classes but often requires additional descriptive information about them [1].”
Generative Adversarial Text to Image Synthesis (2016, background): “Traditionally this type of detailed visual information about an object has been captured in attribute representations distinguishing characteristics the object category encoded into a vector (Farhadi et al., 2009; Kumar et al., 2009; Parikh & Grauman, 2011; Lampert et al., 2014), in particular to en”
ELEVATER: A Benchmark and Toolkit for Evaluating Language-Augmented Visual Models (2022, neutral): “They are usually in single domain, with a manual split of categories to product disjoint training and test categories, e.g., Animal with Attributes (AwA) [31], Caltech-UCSD Birds-200 (CUB) [64], SUN [47], aPY [14], and ZS-ImageNet [52, 17].”
Generating Visual Explanations (2016, background): “Most fine-grained zero-shot and few-shot image classification systems use attributes [23] as auxiliary information.”
iCaRL: Incremental Classifier and Representation Learning (2016, background): “[18], can classify examples of previously unseen classes, but it does not include a training step for those.”
The More You Know: Using Knowledge Graphs for Image Classification (2016, background): “Our work is also related to attribute approaches [8] to vision such as [16] which uses a ﬁxed set of binary attributes to do zero-shot prediction, [33] which uses attributes shared across categories to prevent semantic drift in semi-supervised learning and [5] which automatically discovers…”
Learning to Compare: Relation Network for Few-Shot Learning (2017, methodology): “Datasets Two widely used ZSL benchmarks are selected: AwA (Animals with Attributes) [22] consists of 30,745 images of 50 classes of animals.”
Transductive Multi-View Zero-Shot Learning (2015, methodology): “They can be roughly categorised according to the semantic representation(s) used: DAP and IAP ([27], [28]), M2LATM [15], ALE [1], [36] and [50] use attributes only; HLE/AHLE [1] and Mo/Ma/O/D [38] use both attributes and linguistic knowledge bases (same as us); [54] uses attribute and some…”
Efficient Lifelong Learning with A-GEM (2018, background): “, 2017), and it is also a key ingredient in all the zero/few-shot learning algorithms (Lampert et al., 2014; Xian et al., 2018; Elhoseiny et al., 2017; Wah et al., 2011; Lampert et al., 2009).”
A PAC-Bayesian bound for Lifelong Learning (2013, methodology): “We optimize (17), and due to the squared loss, the problem has a closed form solution: To make results comparable with the baseline algorithms, we report the squared error multiplied by the squared value of the maximum examination score.”
Deep Metric Learning via Lifted Structured Feature Embedding (2015, methodology): “To make full use of the batch, one key idea is to enhance the mini-batch optimization to use all O(m2) pairs in the batch, instead of O(m) separate pairs.”
Zero-Shot Learning—A Comprehensive Evaluation of the Good, the Bad and the Ugly (2017, methodology): “Although Direct Attribute Prediction (DAP) [1] and Indirect At-tribute Prediction (IAP) [1] have been shown to perform poorly compared to compatibility learning frameworks [30], we include them to our evaluation for being historically the most widely used methods in the literature.”
Zero-Shot Learning — The Good, the Bad and the Ugly (2017, methodology): “In zero-shot learning setting test and training class sets are disjoint [17, 21, 22, 29, 40] which can be tackled by solving related sub-problems, e.”
An embarrassingly simple approach to zero-shot learning (2015, background): “Together matrices Y and S provide enough information so that one can obtain the ground truth attributes for each instance.”
Evaluation of output embeddings for fine-grained image classification (2014, methodology): “caption generation [28, 39], face recognition [48, 6], image retrieval [49, 11], action recognition [33, 57] and image classification [29, 2].”