English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Facial expression plays an important role in understanding human emotions.Most recently, deep learning based methods have shown promising for facialexpression recognition. However, the performance of the currentstate-of-the-art facial expression recognition (FER) approaches is directlyrelated to the labeled data for training. To solve this issue, prior worksemploy the pretrain-and-finetune strategy, i.e., utilize a large amount ofunlabeled data to pretrain the network and then finetune it by the labeleddata. As the labeled data is in a small amount, the final network performanceis still restricted. From a different perspective, we propose to performomni-supervised learning to directly exploit reliable samples in a large amountof unlabeled data for network training. Particularly, a new dataset is firstlyconstructed using a primitive model trained on a small number of labeledsamples to select samples with high confidence scores from a face dataset,i.e., MS-Celeb-1M, based on feature-wise similarity. We experimentally verifythat the new dataset created in such an omni-supervised manner cansignificantly improve the generalization ability of the learned FER model.However, as the number of training samples grows, computational cost andtraining time increase dramatically. To tackle this, we propose to apply adataset distillation strategy to compress the created dataset into severalinformative class-wise images, significantly improving the training efficiency.We have conducted extensive experiments on widely used benchmarks, whereconsistent performance gains can be achieved under various settings using theproposed framework. More importantly, the distilled dataset has shown itscapabilities of boosting the performance of FER with negligible additionalcomputational costs.

Omni-supervised Facial Expression Recognition via Distilled Data