COCO18-DensePose-BUPT-PRIV

Size: px

Start display at page:

Download "COCO18-DensePose-BUPT-PRIV"

卓丑籍
5 years ago
Views:

2 * Beijing University of Posts and Telecommunications (BUPT)

3 COCO 2018 DensePose Test AP AP50 AP75 APm APl BUPT-PRIV (Ours) PlumSix ML_Lab Sound of silent DensePose ResNeXt

5 Bottom-up Backbone Keypoints Dense Pose Person Part Grouping Backbone Instance B-Box Instancelevel task Top-down

6 Box head Backbone Mask head Mask R-CNN RoI Align Keypoint head

7 The outline of the human body is not good enough.

8 感谢您下载包图网平台上提供的PPT作品为了您和包图网以及原创作者的利益请勿复制传播销售否则将承担法律责任包图网将对作品进行维权按照传播下载次数进行十倍的索取赔偿 Dense Pose Estimation Person Part Segmentation Can the Mask R-CNN handle the problem of person instance analysis well?

9 Box head Backbone Mask head Mask R-CNN RoI Align Keypoint head Parsing R-CNN DensePose head MR-Parsing Head Parsing head

10 Large scale instance More detail information use P2 for parsing AP P2-P6 (baseline) 48.9 P2-only P2 for parsing Use P2-P6 for RPN, and only P2 for MR-Parsing Head

11 P2 RoI Align 75*125 14*14 250* * *800 Lost too many details!

12 512*32*32 512*32*32 512*32*32 enlarging the roi size AP 14*14 (Original size ) *32 (Ours) *48 (Ours) conv layers Use 32 * 32 roi size and output size is 128 * 128

13 512*32*32 512*32*32 8 conv layers RF = 17 < 32 = roi size From DeepLab V3

use aspp module for paring AP 8 conv layers 0.523 52.2 ppmaspp module module (PSPNet) 0.526 52.9 +0.3 +0.

14 use aspp module for paring AP 8 conv layers ppmaspp module module (PSPNet) ppm aspp module module (PSPNET) From DeepLab V3 Use aspp module to instead of 8 conv layers

15 512*32*32 512*32*32 ASPP add 4 conv layers after aspp AP aspp module conv before aspp conv after aspp both Before ASPP Semantic space transformation? After ASPP High-level semantic feature? Use 4 conv layers after aspp module

16 aspp Small testing scale and less roi for speeding up AP ResNet50 (baseline) MR-Parsing Head scale test scale & 100 rois test oi Align output 4 conv layers A good designed head is very important for instance-level parsing

17 70 COCO 2018 DensePose Val ResNet50 (baseline) + MR-Parsing Head + ResNeXt101 + COCO pre-train + Test aug We get 64.1 map on the test set with single model.

18 感谢您下载包图网平台上提供的PPT作品为了您和包图网以及原创作者的利益请勿复制传播销售否则将承担法律责任包图网将对作品进行维权按照传播下载次数进行十倍的索取赔偿 DensePose ResNeXt101 Parsing R-CNN (ours)

19 What s the next?

20 Box head Backbone MR-Parsing Head RoI Adaptiv Align?? RoI Align Parsing R-CNN

21 And One More Thing

22 We get74.9 AP on the test set!!!

23 Mask Index Index Refine * =

24 感谢您下载包图网平台上提供的PPT作品为了您和包图网以及原创作者的利益请勿复制传播销售否则将承担法律责任包图网将对作品进行维权按照传播下载次数进行十倍的索取赔偿

25 AP (GPS) AP (GPS without Index refine) DensePose ResNet DensePose ResNeXt Parsing R-CNN ResNeXt Parsing R-CNN ResNeXt101 +Test Aug COCO 2018 DensePose Val Without index refine, we can get unreasonabe high scores.

26 GPS jm = GPS j * IoU j mask = GPS j *! "#! %&'! "#! %&', I is class-agnostic index AP (GPS) AP (GPS without Index refine) AP (GPS M ) AP (GPS M without Index refine) (GPS - GPS M ) DensePose ResNet DensePose ResNeXt Parsing R-CNN ResNeXt Parsing R-CNN ResNeXt101 +Test Aug COCO 2018 DensePose Val The proposed new metric GPS M

Mirko_Wang_TrinityForum2018

Mirko_Wang_TrinityForum2018 Travel Retail in the Eyes of Jessica Who we are Attracting Chinese travelers 感谢您下载包图网平台上提供的 PPT作品为了您和包图网以及原创作者的利益请勿复制传播销售否则将承担法律责任包图网将对作品进行维权按