papers AI Learner
The Github is limit! Click to go to the new site.

Pooling Pyramid Network for Object Detection

2018-07-09
Pengchong Jin, Vivek Rathod, Xiangxin Zhu

Abstract

We’d like to share a simple tweak of Single Shot Multibox Detector (SSD) family of detectors, which is effective in reducing model size while maintaining the same quality. We share box predictors across all scales, and replace convolution between scales with max pooling. This has two advantages over vanilla SSD: (1) it avoids score miscalibration across scales; (2) the shared predictor sees the training data over all scales. Since we reduce the number of predictors to one, and trim all convolutions between them, model size is significantly smaller. We empirically show that these changes do not hurt model quality compared to vanilla SSD.

Abstract (translated by Google)
URL

https://arxiv.org/abs/1807.03284

PDF

https://arxiv.org/pdf/1807.03284


Similar Posts

Comments