16 天前

金字塔场景解析网络

金字塔场景解析网络

摘要

场景解析在开放词汇且场景多样化的无限制环境下极具挑战性。本文通过提出金字塔场景解析网络(PSPNet),结合基于多区域的上下文聚合机制,利用金字塔池化模块挖掘全局上下文信息,充分发挥其优势。所提出的全局先验表示方法在场景解析任务中表现出色,能够生成高质量的分割结果;而PSPNet则为像素级预测任务提供了一个优越的框架。该方法在多个公开数据集上均取得了当前最优性能,在2016年ImageNet场景解析挑战赛、PASCAL VOC 2012基准测试以及Cityscapes基准测试中均获得第一名。仅使用单一PSPNet模型,就在PASCAL VOC 2012上取得了85.4%的mIoU(平均交并比)新纪录,在Cityscapes上达到了80.2%的准确率新纪录。

代码仓库

leemathew1998/GradientWeight
pytorch
GitHub 中提及
manideep2510/eye-in-the-sky
tf
GitHub 中提及
qubvel/segmentation_models
tf
GitHub 中提及
PRBonn/bonnet
tf
GitHub 中提及
kazuto1011/pspnet-pytorch
pytorch
GitHub 中提及
daveboat/spp
pytorch
GitHub 中提及
tejaswigowda/semseg-pytorch
pytorch
GitHub 中提及
HenonBamboo/PSPNet-MindSpore
mindspore
GitHub 中提及
intelligent-vehicles/bevdriver
pytorch
GitHub 中提及
kdhingra307/temp
pytorch
GitHub 中提及
UBC-CIC/COVID19-L3-Net
pytorch
GitHub 中提及
Rintarooo/PSPNet
pytorch
GitHub 中提及
leemathew1998/RG
pytorch
GitHub 中提及
AnirudhAchal/Human-Parsing
pytorch
GitHub 中提及
cioppaanthony/rt-sbs
pytorch
GitHub 中提及
tensorflow/models
tf
GitHub 中提及
Lxrd-AJ/Advanced_ML
pytorch
GitHub 中提及
osmr/imgclsmob
mxnet
GitHub 中提及
hszhao/PSPNet
官方
pytorch
GitHub 中提及
EvaBr/AnatomyNets
pytorch
GitHub 中提及
Rosie-Brigham/sesmeg
pytorch
GitHub 中提及
kannyjyk/Nested-UNet
tf
GitHub 中提及
DiMarzioRock7/PSPNet
pytorch
GitHub 中提及
PhanTom2003/PSPnet
pytorch
GitHub 中提及
RituYadav92/Image-segmentation
pytorch
GitHub 中提及
Mind23-2/MindCode-63
mindspore
GitHub 中提及
jqueguiner/camembert-as-a-service
pytorch
GitHub 中提及
yangyucheng000/PSPNet
mindspore
GitHub 中提及
2023-MindSpore-1/ms-code-46
mindspore
GitHub 中提及
163GitHub/AI
pytorch
GitHub 中提及
kingcong/PSPNet
mindspore
GitHub 中提及
fenglian425/Agriculture_AI
pytorch
GitHub 中提及
geekswaroop/Human-Parsing
pytorch
GitHub 中提及

基准测试

基准方法指标
dichotomous-image-segmentation-on-dis-te1PSPNet
E-measure: 0.791
HCE: 267
MAE: 0.089
S-Measure: 0.725
max F-Measure: 0.645
weighted F-measure: 0.557
dichotomous-image-segmentation-on-dis-te2PSPNet
E-measure: 0.828
HCE: 586
MAE: 0.092
S-Measure: 0.763
max F-Measure: 0.724
weighted F-measure: 0.636
dichotomous-image-segmentation-on-dis-te3PSPNet
E-measure: 0.843
HCE: 1111
MAE: 0.092
S-Measure: 0.774
max F-Measure: 0.747
weighted F-measure: 0.657
dichotomous-image-segmentation-on-dis-te4PSPNet
E-measure: 0.815
HCE: 3806
MAE: 0.107
S-Measure: 0.758
max F-Measure: 0.725
weighted F-measure: 0.630
dichotomous-image-segmentation-on-dis-vdPSPNet
E-measure: 0.802
HCE: 1588
MAE: 0.102
S-Measure: 0.744
max F-Measure: 0.691
weighted F-measure: 0.603
lesion-segmentation-on-anatomical-tracings-of-1PSPNet
Dice: 0.3571
IoU: 0.254
Precision: 0.4769
Recall: 0.3335
real-time-semantic-segmentation-on-camvidPSPNet
Frame (fps): 5.4
Time (ms): 185.0
real-time-semantic-segmentation-on-nyu-depth-1PSPNet101
Speed(ms/f): 72
mIoU: 43.2
real-time-semantic-segmentation-on-nyu-depth-1PSPNet50
Speed(ms/f): 47
mIoU: 41.8
real-time-semantic-segmentation-on-nyu-depth-1PSPNet18
Speed(ms/f): 19
mIoU: 35.9
semantic-segmentation-on-ade20kPSPNet (ResNet-101)
Validation mIoU: 43.29
semantic-segmentation-on-ade20kPSPNet (ResNet-152)
Validation mIoU: 43.51
semantic-segmentation-on-ade20kPSPNet
Test Score: 55.38
Validation mIoU: 44.94
semantic-segmentation-on-ade20k-valPSPNet (ResNet-101)
mIoU: 43.29%
semantic-segmentation-on-ade20k-valPSPNet (ResNet-152)
mIoU: 43.51%
semantic-segmentation-on-bdd100k-valPSPNet
mIoU: 62.3
semantic-segmentation-on-cityscapesPSPNet
Mean IoU (class): 78.4%
semantic-segmentation-on-cityscapesPSPNet++
Mean IoU (class): 80.2%
semantic-segmentation-on-cityscapes-valPSPNet (Dilated-ResNet-101)
mIoU: 79.7
semantic-segmentation-on-dada-segPSPNet (ResNet-101)
mIoU: 20.1
semantic-segmentation-on-densepassPSPNet (ResNet-50)
mIoU: 29.5%
semantic-segmentation-on-pascal-contextPSPNet (ResNet-101)
mIoU: 47.8
semantic-segmentation-on-pascal-voc-2012PSPNet
Mean IoU: 85.4%
semantic-segmentation-on-pascal-voc-2012PSPNet (ResNet-101)
Mean IoU: 82.6%
semantic-segmentation-on-potsdamPSPNet
mIoU: 82.98
semantic-segmentation-on-scannetv2PSPNet
Mean IoU: 47.5%
semantic-segmentation-on-selmaPSPNet
mIoU: 68.4
semantic-segmentation-on-trans10kPSPNet
GFLOPs: 187.03
mIoU: 68.23%
semantic-segmentation-on-urbanlfPSPNet
mIoU (Real): 76.34
mIoU (Syn): 75.78
semantic-segmentation-on-us3dPSNet
mIoU: 73.12
semantic-segmentation-on-vaihingenPSPNet
mIoU: 76.79
thermal-image-segmentation-on-mfn-datasetPSPNet
mIOU: 46.1
video-semantic-segmentation-on-camvidPSPNet-50
Mean IoU: 76
video-semantic-segmentation-on-cityscapes-valPSPNet-101 [20]
mIoU: 79.7
video-semantic-segmentation-on-cityscapes-valPSPNet-50 [20]
mIoU: 78.1

用 AI 构建 AI

从想法到上线——通过免费 AI 协同编程、开箱即用的环境和市场最优价格的 GPU 加速您的 AI 开发

AI 协同编程
即用型 GPU
最优价格
立即开始

Hyper Newsletters

订阅我们的最新资讯
我们会在北京时间 每周一的上午九点 向您的邮箱投递本周内的最新更新
邮件发送服务由 MailChimp 提供
金字塔场景解析网络 | 论文 | HyperAI超神经