SNIPER: Efficient Multi-Scale Training

We present SNIPER, an algorithm for performing efficient multi-scale training in instance level visual recognition tasks. Instead of processing every pixel in an image pyramid, SNIPER processes context regions around ground-truth instances (referred to as chips) at the appropriate scale... (read more)

PDF Abstract NeurIPS 2018 PDF NeurIPS 2018 Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Object Detection COCO test-dev SNIPER (ResNet-101) box AP 46.1 # 28
AP50 67.0 # 25
AP75 51.6 # 28
APS 29.6 # 26
APM 48.9 # 31
APL 58.1 # 29
Object Detection COCO test-dev SNIPER (ResNet-50) box AP 43.5 # 41
AP50 65.0 # 32
AP75 48.6 # 40
APS 26.1 # 42
APM 46.3 # 48
APL 56.0 # 40
Object Detection PASCAL VOC 2007 SNIPER MAP 86.9% # 1

Methods used in the Paper