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

Knowledge Distillation (KD) utilizes training data as a transfer set totransfer knowledge from a complex network (Teacher) to a smaller network(Student). Several works have recently identified many scenarios where thetraining data may not be available due to data privacy or sensitivity concernsand have proposed solutions under this restrictive constraint for theclassification task. Unlike existing works, we, for the first time, solve amuch more challenging problem, i.e., "KD for object detection with zeroknowledge about the training data and its statistics". Our proposed approachprepares pseudo-targets and synthesizes corresponding samples (termed as"Multi-Object Impressions"), using only the pretrained Faster RCNN Teachernetwork. We use this pseudo-dataset as a transfer set to conduct zero-shot KDfor object detection. We demonstrate the efficacy of our proposed methodthrough several ablations and extensive experiments on benchmark datasets likeKITTI, Pascal and COCO. Our approach with no training samples, achieves arespectable mAP of 64.2% and 55.5% on the student with same and half capacitywhile performing distillation from a Resnet-18 Teacher of 73.3% mAP on KITTI.

Beyond Classification: Knowledge Distillation using Multi-Object Impressions