The results show that class disjointness alone is insufficient to remove the influence of target-domain data, highlighting the need to move beyond in-domain pre-training as the default few-shot evaluation protocol, since it can overestimate performance in realistic scenarios where target-domain data are scarce.
Abstract
Few-shot learning is commonly evaluated under protocols that pre-train a model on a large auxiliary set whose classes are disjoint from the target episodes yet drawn from the same visual domain. This paper examines whether such protocols truly reflect low-data learning. We systematically compare no pre-training, class-disjoint in-domain pre-training, supervised out-of-domain pre-training, and label-free out-of-domain pre-training across eight datasets, three few-shot architectures, and multiple way-shot settings. Our results show that class disjointness alone is insufficient to remove the influence of target-domain data. In-domain pre-training improves over no pre-training by 33.41 percentage points on average, whereas supervised out-of-domain pre-training yields 23.75 percentage points, revealing a 9.66-point optimistic bias associated with domain overlap. Although out-of-domain pre-training is more realistic in applications where target-domain data are scarce, its effectiveness depends strongly on the compatibility between source and target domains. We further show that labeled source data are not strictly required, with an augmentation-based label-free strategy reaching an average gain of 27.71 percentage points and closely matching supervised out-of-domain pre-training at 27.97 percentage points. Finally, we introduce a descriptor-based source-selection strategy that estimates source-domain suitability before pre-training, reaching a median gap of only 1.37 percentage points to oracle selection. These findings highlight the need to move beyond in-domain pre-training as the default few-shot evaluation protocol, since it can overestimate performance in realistic scenarios where target-domain data are scarce.
Few-shot image classification (FSIC) addresses image recognition when only a few labeled examples are available for each novel class. This survey reviews five major FSIC method families: meta-learning, transfer learning, data augmentation, attribute-based methods, and vision-language model adaptation. The $N$-way $K$-s...
A. Daneshpour, Stefan Bucz, D. Jantassova et al.· Eurasian journal of mathemat...· 0 citations
Spectral Transductive Refinement (STR), a training-free transductive inference rule that exploits the geometry of the complete support-query episode, operating entirely at inference time, attains the highest 1-shot average among compared methods and remains competitive at 5-shot, rivalling approaches relying on heavy s...
F. Rahman, S. Rohan, Mahmound Sayed et al.· arXiv.org· 0 citations
Across ResNet12 and Conv64 in 5-way 1-shot and 5-shot evaluation, SAMPLESELECT gives the best OOD accuracy among the compared methods and improves the matched full-representation control by 4.90-8.38 percentage points.
Feng-Rui Liu, Ning Shen, Yi Li et al.· arXiv.org· 0 citations
Vision-language models such as CLIP achieve strong zero-shot classification, yet under distribution shift, visual embeddings drift from fixed text embeddings. Training-free calibration avoids the per-sample optimization of prompt learning, but prior feature calibration gives each image the full bias of one hard cluster...
A unified comparison of attribute-based ZSL, episodic meta-learning, metric and prototype estimators, graph and optimal-transport inference, generative any-shot models, and adaptation of vision–language models is analyzed.
Jie Li, Yu-Bo Sun, Xun Du et al.· Mathematics· 0 citations
Synthetic hard negatives generated in the representation space have proven effective for unimodal self-supervised learning, but transferring this idea to vision-language pretraining is not straightforward. We analyze six representation-space synthesis strategies and identify two failure modes in their transfer to visio...
Nikolaos Giakoumoglou, Paschalis Giakoumoglou, Andreas Floros et al.· 0 citations
Professor Paris Smaragdis develops computers with superhuman hearing for applications like speech recognition, music understanding, and environmental monitoring.
MIT News · Artificial Intelligence· news.mit.eduOct 13, 2026
Machine-learning networks can learn without detailed instructions from a designer. Those that do may offer insight into how the brain reorganizes itself during learning.
Exploring how generative AI could make machine vision more accessible to businesses. The post GenEye in a Box: Making Machine Vision Something You Can Just Ask For appeared first on GPT-Lab.
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.