Partner-Assisted Learning for Few-Shot Image Classification

Ma, Jiawei; Hanchen, Xie,; Han, Guangxing; Chang, Shih-Fu; Galstyan, Aram; AbdAlmageed, Wael

doi:10.1109/iccv48922.2021.01040

Cited by 50 publications

(27 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…SimpleShot [29] 62.85 ± 0.20 80.02 ± 0.14 Baseline++ [30] 53.97 ± 0.79 75.90 ± 0.61 TADAM [35] 58.50 ± 0.30 76.70 ± 0.30 ProtoNet [10] 60.37 ± 0.83 78.02 ± 0.57 R2-D2 (+ens) [20] 64.79 ± 0.45 81.08 ± 0.32 FEAT [36] 66.78 82.05 CNL [37] 67.96 ± 0.98 83.36 ± 0.51 MERL [38] 67.40 ± 0.43 83.40 ± 0.28 Deep EMD v2 [13] 68.77 ± 0.29 84.13 ± 0.53 PAL [8] 69.37 ± 0.64 84.40 ± 0.44 inv-equ [39] 67.28 ± 0.80 84.78 ± 0.50 CSEI [40] 68.94 ± 0.28 85.07 ± 0.50 COSOC [9] 69.…”

Section: Discussionmentioning

confidence: 99%

“…Contrastive learning aims to train a model to learn to maximize similarities between transformed instances of the same image and minimize agreement between transformed instances of different images [21], [22], [23], [24], [9]. Supervised contrastive learning is a variant which has been recently used in few-shot learning, where similarity is maximized between instances of a class instead of the same image [25], [8].…”

Section: B Backbone Trainingmentioning

confidence: 99%

“…Distillation has been recently used in the few-shot literature. The idea is to transfer knowledge from a teacher model to a student model by forcing the latter to match the class probabilities distribution of the teacher [26], [27], [8].…”

Section: Exploiting Multiple Backbonesmentioning

confidence: 99%

“…SimpleShot [29] 69.09 ± 0.22 84.58 ± 0.16 ProtoNet [10] 65.65 ± 0.92 83.40 ± 0.65 FEAT [36] 70.80 ± 0.23 84.79 ± 0.16 PAL [8] 72.25 ± 0.72 86.95 ± 0.47 DeepEMD v2 [13] 74.29 ± 0.32 86.98 ± 0.60 MERL [38] 72.14 ± 0.51 87.01 ± 0.35 COSOC [9] 73.57 ± 0.43 87.57 ± 0.10 CNL [37] 73.42 ± 0.95 87.72 ± 0.75 invariance-equivariance [39] 72.21 ± 0.90 87.08 ± 0.58 CSEI [40] 73…”

Section: Table IV 1-shot and 5-shot Accuracy Of State-of-the-art Meth...mentioning

confidence: 99%

“…In recent years, a large number of contributions have introduced methodologies to cope with few-shot problems. There are a lot of ingredients involved, including distillation [8], contrastive learning [9], episodic training [10], mixup [11], manifold mixup [12], [7] and self-supervision [12]. As a consequence, it can appear quite opaque what are the effective ingredients, and whether their performance can be reproduced across different datasets or settings.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

EASY: Ensemble Augmented-Shot Y-shaped Learning: State-Of-The-Art Few-Shot Classification with Simple Ingredients

Bendou¹,

Hu²,

Lafargue³

et al. 2022

Preprint

View full text Add to dashboard Cite

Few-shot learning aims at leveraging knowledge learned by one or more deep learning models, in order to obtain good classification performance on new problems, where only a few labeled samples per class are available. Recent years have seen a fair number of works in the field, introducing methods with numerous ingredients. A frequent problem, though, is the use of suboptimally trained models to extract knowledge, leading to interrogations on whether proposed approaches bring gains compared to using better initial models without the introduced ingredients. In this work, we propose a simple methodology, that reaches or even beats state of the art performance on multiple standardized benchmarks of the field, while adding almost no hyperparameters or parameters to those used for training the initial deep learning models on the generic dataset. This methodology offers a new baseline on which to propose (and fairly compare) new techniques or adapt existing ones.

show abstract