DataTalks #28: Cross-Domain Few-Shot Classification ✝️🔫

DataTalks #28: Cross-Domain Few-Shot Classification ✝️🔫

Our 28th (!) DataTalks meetup will be held online, and will host U.C. Merced Ph.D. student Hung-Yu Tseng, who will present his ICLR 2020 Spotlight paper on Cross-Domain Few-Shot Classification via Learned Feature-Wise Transformation.

𝗭𝗼𝗼𝗺 𝗹𝗶𝗻𝗸: https://us02web.zoom.us/j/81310454908?pwd=bGp1M2ZESnB5TmFkbUNUZ2Z0aU9wdz09

𝗔𝗴𝗲𝗻𝗱𝗮:
🔶 20:00 - 20:40 - Cross-Domain Few-Shot Classification via Learned Feature-Wise Transformation - Hung-Yu Tseng, U.C. Merced

---------------------

𝗖𝗿𝗼𝘀𝘀-𝗗𝗼𝗺𝗮𝗶𝗻 𝗙𝗲𝘄-𝗦𝗵𝗼𝘁 𝗖𝗹𝗮𝘀𝘀𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻 𝘃𝗶𝗮 𝗟𝗲𝗮𝗿𝗻𝗲𝗱 𝗙𝗲𝗮𝘁𝘂𝗿𝗲-𝗪𝗶𝘀𝗲 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝗼𝗻 - 𝗛𝘂𝗻𝗴-𝗬𝘂 𝗧𝘀𝗲𝗻𝗴, 𝗨.𝗖. 𝗠𝗲𝗿𝗰𝗲𝗱

Few-shot classification aims to recognise novel categories with only few labeled images in each class. Existing metric-based few-shot classification algorithms predict categories by comparing the feature embeddings of query images with those from a few labeled images (support examples) using a learned metric function. While promising performance has been demonstrated, these methods often fail to generalize to unseen domains due to large discrepancy of the feature distribution across domains. In this work, we address the problem of few-shot classification under domain shifts for metric-based methods.

Our core idea is to use feature-wise transformation layers for augmenting the image features using affine transforms to simulate various feature distributions under different domains in the training stage. To capture variations of the feature distributions under different domains, we further apply a learning-to-learn approach to search for the hyper-parameters of the feature-wise transformation layers. We conduct extensive experiments and ablation studies under the domain generalization setting using five few-shot classification datasets: mini-ImageNet, CUB, Cars, Places, and Plantae.

Experimental results demonstrate that the proposed feature-wise transformation layer is applicable to various metric-based models, and provides consistent improvements on the few-shot classification performance under domain shift.

𝗣𝗮𝗽𝗲𝗿 𝗹𝗶𝗻𝗸: https://arxiv.org/abs/2001.08735

𝗕𝗶𝗼: Hung-Yu Tseng is a 3-rd year Ph.D. student in the Vision and Learning Lab at U.C. Merced, advised by Prof. Ming-Hsuan Yang.

𝗭𝗼𝗼𝗺 𝗹𝗶𝗻𝗸: https://us02web.zoom.us/j/81310454908?pwd=bGp1M2ZESnB5TmFkbUNUZ2Z0aU9wdz09