configs/ocrnet/README.md

OCRNet

Object-Contextual Representations for Semantic Segmentation

Introduction

Official Repo

Code Snippet

Abstract

In this paper, we address the problem of semantic segmentation and focus on the context aggregation strategy for robust segmentation. Our motivation is that the label of a pixel is the category of the object that the pixel belongs to. We present a simple yet effective approach, object-contextual representations, characterizing a pixel by exploiting the representation of the corresponding object class. First, we construct object regions based on a feature map supervised by the ground-truth segmentation, and then compute the object region representations. Second, we compute the representation similarity between each pixel and each object region, and augment the representation of each pixel with an object contextual representation, which is a weighted aggregation of all the object region representations according to their similarities with the pixel. We empirically demonstrate that the proposed approach achieves competitive performance on six challenging semantic segmentation benchmarks: Cityscapes, ADE20K, LIP, PASCAL VOC 2012, PASCAL-Context and COCO-Stuff. Notably, we achieved the \nth{2} place on the Cityscapes leader-board with a single model.

Results and models

Cityscapes

HRNet backbone

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	Device	mIoU	mIoU(ms+flip)	config	download
OCRNet	HRNetV2p-W18-Small	512x1024	40000	3.5	10.45	A100	76.61	78.01	config	model | log
OCRNet	HRNetV2p-W18	512x1024	40000	4.7	7.50	V100	77.72	79.49	config	model | log
OCRNet	HRNetV2p-W48	512x1024	40000	8	4.22	V100	80.58	81.79	config	model | log
OCRNet	HRNetV2p-W18-Small	512x1024	80000	-	-	V100	77.16	78.66	config	model | log
OCRNet	HRNetV2p-W18	512x1024	80000	-	-	V100	78.57	80.46	config	model | log
OCRNet	HRNetV2p-W48	512x1024	80000	-	-	V100	80.70	81.87	config	model | log
OCRNet	HRNetV2p-W18-Small	512x1024	160000	-	-	V100	78.45	79.97	config	model | log
OCRNet	HRNetV2p-W18	512x1024	160000	-	-	V100	79.47	80.91	config	model | log
OCRNet	HRNetV2p-W48	512x1024	160000	-	-	V100	81.35	82.70	config	model | log

ResNet backbone

Method	Backbone	Crop Size	Batch Size	Lr schd	Mem (GB)	Inf time (fps)	Device	mIoU	mIoU(ms+flip)	config	download
OCRNet	R-101-D8	512x1024	8	40000	-	-	V100	80.09	-	config	model | log
OCRNet	R-101-D8	512x1024	16	40000	8.8	3.02	V100	80.30	-	config	model | log
OCRNet	R-101-D8	512x1024	16	80000	8.8	3.02	V100	80.81	-	config	model | log

ADE20K

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	Device	mIoU	mIoU(ms+flip)	config	download
OCRNet	HRNetV2p-W18-Small	512x512	80000	6.7	28.98	V100	35.06	35.80	config	model | log
OCRNet	HRNetV2p-W18	512x512	80000	7.9	18.93	V100	37.79	39.16	config	model | log
OCRNet	HRNetV2p-W48	512x512	80000	11.2	16.99	V100	43.00	44.30	config	model | log
OCRNet	HRNetV2p-W18-Small	512x512	160000	-	-	V100	37.19	38.40	config	model | log
OCRNet	HRNetV2p-W18	512x512	160000	-	-	V100	39.32	40.80	config	model | log
OCRNet	HRNetV2p-W48	512x512	160000	-	-	V100	43.25	44.88	config	model | log

Pascal VOC 2012 + Aug

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	Device	mIoU	mIoU(ms+flip)	config	download
OCRNet	HRNetV2p-W18-Small	512x512	20000	3.5	31.55	V100	71.70	73.84	config	model | log
OCRNet	HRNetV2p-W18	512x512	20000	4.7	19.91	V100	74.75	77.11	config	model | log
OCRNet	HRNetV2p-W48	512x512	20000	8.1	17.83	V100	77.72	79.87	config	model | log
OCRNet	HRNetV2p-W18-Small	512x512	40000	-	-	V100	72.76	74.60	config	model | log
OCRNet	HRNetV2p-W18	512x512	40000	-	-	V100	74.98	77.40	config	model | log
OCRNet	HRNetV2p-W48	512x512	40000	-	-	V100	77.14	79.71	config	model | log

Citation

@article{YuanW18,
  title={Ocnet: Object context network for scene parsing},
  author={Yuhui Yuan and Jingdong Wang},
  booktitle={arXiv preprint arXiv:1809.00916},
  year={2018}
}

@article{YuanCW20,
  title={Object-Contextual Representations for Semantic Segmentation},
  author={Yuhui Yuan and Xilin Chen and Jingdong Wang},
  booktitle={ECCV},
  year={2020}
}