pisa

PISA

Prime Sample Attention in Object Detection

Abstract

It is a common paradigm in object detection frameworks to treat all samples equally and target at maximizing the performance on average. In this work, we revisit this paradigm through a careful study on how different samples contribute to the overall performance measured in terms of mAP. Our study suggests that the samples in each mini-batch are neither independent nor equally important, and therefore a better classifier on average does not necessarily mean higher mAP. Motivated by this study, we propose the notion of Prime Samples, those that play a key role in driving the detection performance. We further develop a simple yet effective sampling and learning strategy called PrIme Sample Attention (PISA) that directs the focus of the training process towards such samples. Our experiments demonstrate that it is often more effective to focus on prime samples than hard samples when training a detector. Particularly, On the MSCOCO dataset, PISA outperforms the random sampling baseline and hard mining schemes, e.g., OHEM and Focal Loss, consistently by around 2% on both single-stage and two-stage detectors, even with a strong backbone ResNeXt-101.

Results and Models

PISA	Network	Backbone	Lr schd	box AP	mask AP	Config	Download
×	Faster R-CNN	R-50-FPN	1x	36.4		-
√	Faster R-CNN	R-50-FPN	1x	38.4		config	model | log
×	Faster R-CNN	X101-32x4d-FPN	1x	40.1		-
√	Faster R-CNN	X101-32x4d-FPN	1x	41.9		config	model | log
×	Mask R-CNN	R-50-FPN	1x	37.3	34.2	-
√	Mask R-CNN	R-50-FPN	1x	39.1	35.2	config	model | log
×	Mask R-CNN	X101-32x4d-FPN	1x	41.1	37.1	-
√	Mask R-CNN	X101-32x4d-FPN	1x
×	RetinaNet	R-50-FPN	1x	35.6		-
√	RetinaNet	R-50-FPN	1x	36.9		config	model | log
×	RetinaNet	X101-32x4d-FPN	1x	39.0		-
√	RetinaNet	X101-32x4d-FPN	1x	40.7		config	model | log
×	SSD300	VGG16	1x	25.6		-
√	SSD300	VGG16	1x	27.6		config	model | log
×	SSD300	VGG16	1x	29.3		-
√	SSD300	VGG16	1x	31.8		config	model | log

Notes:

• In the original paper, all models are trained and tested on mmdet v1.x, thus results may not be exactly the same with this release on v2.0.
• It is noted PISA only modifies the training pipeline so the inference time remains the same with the baseline.

Citation

@inproceedings{cao2019prime,
  title={Prime sample attention in object detection},
  author={Cao, Yuhang and Chen, Kai and Loy, Chen Change and Lin, Dahua},
  booktitle={IEEE Conference on Computer Vision and Pattern Recognition},
  year={2020}
}