Few-Shot Image Recognition by Predicting Parameters from Activations

By Siyuan Qiao, Chenxi Liu, Wei Shen, Alan L. Yuille. 2017

589 citations recorded by the source corpus.

In this paper, we are interested in the few-shot learning problem. In particular, we focus on a challenging scenario where the number of categories is large and the number of examples per novel category is very limited, e.g. 1, 2, or 3. Motivated by the close relationship between the parameters and the activations in a neural network associated with the same category, we propose a novel method that can adapt a pre-trained neural network to novel categories by directly predicting the parameters from the activations. Zero training is required in adaptation to novel categories, and fast inference is realized by a single forward pass. We evaluate our method by doing few-shot image recognition on the ImageNet dataset, which achieves the state-of-the-art classification accuracy on novel categories by a significant margin while keeping comparable performance on the large-scale categories. We also test our method on the MiniImageNet dataset and it strongly outperforms the previous state-of-the-art methods.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging, Representation & Self-Supervised Learning

How researchers see this paper

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

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Radiology, Nuclear Medicine and Imaging

In Representation & Self-Supervised Learning

Citations in the corpus

Weight Standardization (2019, background): “Deep neural networks achieve state-of-the-arts in many vision tasks [4, 14, 20, 25, 31, 32, 39, 46].”
Rethinking Few-Shot Image Classification: a Good Embedding Is All You Need? (2020, neutral): “We notice that previous works [38, 44, 34, 47] have also leveraged the standard cross-entropy pre-training on the meta-training set.”
Fast and Flexible Multi-Task Classification Using Conditional Neural Adaptive Processes (2019, background): “Conversely, function approximators can be used to directly map training data to the desired parameters (we refer to this as amortization) [5, 10].”
Meta-Learning Probabilistic Inference for Prediction (2018, methodology): “…et al., 2018) to cover a broader class of methods, including gradient based meta-learning (Finn et al., 2017; Ravi and Larochelle, 2017), metric based meta-learning (Snell et al., 2017), amortized MAP inference (Qiao et al., 2017) and conditional probability modelling (Garnelo et al., 2018a;b).”
Meta-learning with differentiable closed-form solvers (2018, methodology): “For our comparison, we report the results of methods which train their models from scratch for few-shot classiﬁcation, omitting very recent work of Qiao et al. (2018) and Gidaris & Komodakis (2018), which instead make use of pre-trained embeddings.”
Meta-Learning with Latent Embedding Optimization (2018, methodology): “30% (Qiao et al., 2017) 59 (Ren et al., 2018) 53 .”
Diversity With Cooperation: Ensemble Methods for Few-Shot Classification (2019, methodology): “ing CNNs from scratch using meta-learning performs substantially worse than if CNN features are trained in a standard fashion—that is, by minimizing a classical loss function relying on corpus annotations; on the other hand, learning only the last layer with meta-learning has been found to produce b”
Discriminative k-shot learning using probabilistic models (2017, methodology): “The network architecture is inspired by the ResNet-34 architecture for ImageNet [10] that uses convolution blocks, with two convolutions each, that are bridged by skip connections.”
Few-Shot and Continual Learning with Attentive Independent Mechanisms (2021, methodology): “We follow the setup in [20, 15, 40, 14] by using a 4-layer convolutional network with 64 feature channels (Conv-4-64) or a WideResNet (WRN-2810) [51] as our feature extractor, fψ .”
Meta-Transfer Learning for Few-Shot Learning (2018, methodology): “Some few-shot learning methods have been proposed to use pre-trained weights as initialization [20, 29, 37, 44, 40].”
Defining Benchmarks for Continual Few-Shot Learning (2020, background): “Related to continual few-shot learning is the field of incremental few-shot learning (Qiao et al., 2018; Gidaris & Komodakis, 2018).”
Meta-Learning in Neural Networks: A Survey (2020, background): “works [14], [51], convolutional networks [39] or hypernetworks [81], [82] that embed training instances and labels of”
A Baseline for Few-Shot Image Classification (2019, methodology): “(2)For instance, [1, 7] use different versions of Mini-ImageNet; [3] report results for a backbone pre-trained on the training set while [8] use both the training and validation sets; [9] use full-sized images from the parent Imagenet-1k dataset [6]; [10, 11, 12, 3, 2] all use different architecture”
Learning Adaptive Classifiers Synthesis for Generalized Few-Shot Learning (2019, methodology): “plied to the meta-training set of the TieredImageNet, Heterogeneous, and OceHome datasets, where a 351-way, 100-way, and 25-way classiers are pre-trained. A.2 Feature Network Specication We follow [45, 52] when investigating the multi-domain GFSL, where images are resized to 84 84 3. In concrete wor”
Meta R-CNN: Towards General Solver for Instance-Level Low-Shot Learning (2019, background): “Distinct from them, meta-learning [40, 37, 12, 32, 3, 16, 43, 11] designs to learn a meta-learner to parametrize the optimization algorithm or predict the parameters of a classifier, so-called “learning-to-learn”.”
Meta-Curvature (2019, methodology): “In order to overcome the nature of few-shot learning, the generative models have been suggested to augment the training data [40, 45] or generate model parameters for the speciﬁed task [39, 33].”
Boosting Few-Shot Visual Learning With Self-Supervision (2019, neutral): “, training images, of a few-shot task [26, 33, 53, 55, 57, 59]; methods learning to map a test example to a class label by accessing memory modules that store training examples for that task [11, 25, 35, 38, 50]; approaches that learn how to generate the weights of a classifier [12, 16, 43, 44, 13] ”
Generalizing from a Few Examples (2019, methodology): “milar features taken from the novel class and reusing their classifier parameters. Only the threshold for classification is adjusted to avoid confusion with those similar classes. Similar to [89] and [90], a pre-trained CNN is adapted to deal with new class in [40]. But instead of solely using embed”