Low-shot visual object recognition

By Bharath Hariharan, Ross B. Girshick. 2016

51 citations recorded by the source corpus.

Low-shot visual learning - the ability to recognize novel object categories from very few examples - is a hallmark of human visual intelligence. Existing machine learning approaches fail to generalize in the same way. To make progress on this foundational problem, we present a novel protocol to evaluate low-shot learning on complex images where the learner is permitted to first build a feature representation. Then, we propose and evaluate representation regularization techniques that improve the effectiveness of convolutional networks at the task of low-shot learning, leading to a 2x reduction in the amount of training data required at equal accuracy rates on the challenging ImageNet dataset.

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

Growing a Brain: Fine-Tuning by Increasing Model Capacity (2017, background): “Transfer learning is related to both multi-task learning [32, 4, 28, 10, 11, 45, 24, 2] and learning novel categories from few examples [49, 19, 21, 35, 51, 22, 5, 13, 50, 47, 31].”
Recasting Gradient-Based Meta-Learning as Hierarchical Bayes (2018, background): “…de-sign of network architectures that ingest the few-shot training samples directly ( e.g., Koch, 2015; Vinyals et al., 2016; Snell et al., 2017; Hariharan & Girshick, 2017; Triantaﬁllou et al., 2017), or formulating the problem as one of learning to learn , or meta-learning ( e.g.,…”
Deep Meta-Learning: Learning to Learn in the Concept Space (2018, background): “(Hariharan & Girshick, 2016) generate dummy examples for the task of interest by using the geometric transformations inferred from existing examples of other categories.”
ChestX-Ray8: Hospital-Scale Chest X-Ray Database and Benchmarks on Weakly-Supervised Classification and Localization of Common Thorax Diseases (2017, background): “Furthermore, it could also be adopted for one/low-shot learning setup [15], in which only one or several samples are needed to initialize the learning and the system will then evolve by itself with more unlabeled data.”
Generative Adversarial Residual Pairwise Networks for One Shot Learning (2017, unknown): “Another work which uses generated images for improving one shot learning is [11], although they train their model as a transformer for test time generations using cosine similarity in the training set to generate analogies instead of an adversarial loss.”
One-Shot Learning for Semantic Segmentation (2017, methodology): “Discriminative methods in which the parameters of the base classiﬁer (learned on training classes) are adapted to the new class [1, 2, 14, 40] are closely related to our work.”
One-Shot Learning in Discriminative Neural Networks (2017, background): “Other research has explored whether ‘low-shot’ learning can be achieved by modifications to the learning objective of a standard ConvNet architecture [Hariharan and Girshick, 2016].”
Learning to Learn: Model Regression Networks for Easy Small Sample Learning (2016, background): “In particular, inter-class transfer [40] and crossgeneralization [7] are achieved by discovering shared feature representations: 1) captured by linear or nonlinear feature transformations [14, 63, 85, 31, 1, 94, 48], 2) obtained by feature selection [59, 60, 27] or regularization [37], 3) described ”
Learning to Detect Salient Objects with Image-Level Supervision (2017, background): “Note that the regularization term in (6) is imposed on the feature representations rather than the weight parameters, which is reminiscent of a recent work [12], where L1 regularization on feature is used to enforce better generalization from low-shot learning.”
Learning from Small Sample Sets by Combining Unsupervised Meta-Training with CNNs (2016, background): “Such transferability is promising but still restrictive, especially for novel-category recognition from few examples [11, 12, 13, 14, 15, 16, 17, 18].”
LCNN: Lookup-Based Convolutional Neural Network (2016, background): “Learning from few examples per category explored by [17].”
ProtoGAN: Towards Few Shot Learning for Action Recognition (2019, methodology): “Few-shot learning problem can be classified into two broad settings [8] based on their evaluation protocols: standard FSL (FSL) and the more realistic Generalized-FSL (GFSL).”