Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of...

61
Gang Yu 旷视研究院 Object Detection in Recent 3 Years Beyond RetinaNet and Mask R-CNN

Transcript of Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of...

Page 1: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Gang Yu

旷视研究院

Object Detection in Recent 3 YearsBeyond RetinaNet and Mask R-CNN

Page 2: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Schedule of Tutorial

• Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu)• Lecture 2: AutoML for Object Detection (Xiangyu Zhang)• Lecture 3: Finegrained Visual Analysis (Xiu-shen Wei)

Page 3: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Outline

• Introduction to Object Detection• Modern Object detectors

• One Stage detector vs Two-stage detector• Challenges

• Backbone• Head• Pretraining• Scale• Batch Size• Crowd• NAS• Fine-Grained

• Conclusion

Page 4: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Outline

• Introduction to Object Detection• Modern Object detectors

• One Stage detector vs Two-stage detector• Challenges

• Backbone• Head• Pretraining• Scale• Batch Size• Crowd• NAS• Fine-Grained

• Conclusion

Page 5: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

What is object detection?

Page 6: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

What is object detection?

Page 7: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Detection - Evaluation Criteria

Average Precision (AP) and mAP

Figures are from wikipedia

Page 8: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Detection - Evaluation Criteria

mmAP

Figures are from http://cocodataset.org

Page 9: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

How to perform a detection?

• Sliding window: enumerate all the windows (up to millions of windows)• VJ detector: cascade chain

• Fully Convolutional network• shared computation

Robust Real-time Object Detection; Viola, Jones; IJCV 2001

http://www.vision.caltech.edu/html-files/EE148-2005-Spring/pprs/viola04ijcv.pdf

Page 10: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

General Detection Before Deep Learning

• Feature + classifier• Feature

• Haar Feature• HOG (Histogram of Gradient)• LBP (Local Binary Pattern)• ACF (Aggregated Channel Feature)• …

• Classifier• SVM• Bootsing• Random Forest

Page 11: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Traditional Hand-crafted Feature: HoG

Page 12: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Traditional Hand-crafted Feature: HoG

Page 13: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

General Detection Before Deep Learning

Traditional Methods

• Pros• Efficient to compute (e.g., HAAR, ACF) on CPU• Easy to debug, analyze the bad cases• reasonable performance on limited training data

• Cons• Limited performance on large dataset• Hard to be accelerated by GPU

Page 14: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Deep Learning for Object Detection

Based on the whether following the “proposal and refine”

• One Stage• Example: Densebox, YOLO (YOLO v2), SSD, Retina Net• Keyword: Anchor, Divide and conquer, loss sampling

• Two Stage• Example: RCNN (Fast RCNN, Faster RCNN), RFCN, FPN,

MaskRCNN• Keyword: speed, performance

Page 15: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

A bit of History

ImageFeature

Extractor

classification

localization

(bbox)

One stage detector

Densebox (2015) UnitBox (2016) EAST (2017)

YOLO (2015) Anchor Free

Anchor importedYOLOv2 (2016)

SSD (2015)

RON(2017)

RetinaNet(2017)

DSSD (2017)

two stages detector

ImageFeature

Extractor

classification

localization

(bbox)

Proposal

classification

localization

(bbox)

Refine

RCNN (2014) Fast RCNN(2015)Faster RCNN (2015)

RFCN (2016)

MultiBox(2014)

RFCN++ (2017)

FPN (2017)

Mask RCNN (2017)

OverFeat(2013)

Page 16: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Outline

• Introduction to Object Detection• Modern Object detectors

• One Stage detector vs Two-stage detector• Challenges

• Backbone• Head• Pretraining• Scale• Batch Size• Crowd• NAS• Fine-Grained

• Conclusion

Page 17: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Modern Object detectors

Backbone Head

• Modern object detectors• RetinaNet

• f1-f7 for backbone, f3-f7 with 4 convs for head• FPN with ROIAlign

• f1-f6 for backbone, two fcs for head• Recall vs localization

• One stage detector: Recall is high but compromising the localization ability• Two stage detector: Strong localization ability

Postprocess

NMS

Page 18: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

One Stage detector: RetinaNet

• FPN Structure

• Focal loss

Focal Loss for Dense Object Detection， Lin etc, ICCV 2017 Best student paper

Page 19: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

One Stage detector: RetinaNet

• FPN Structure

• Focal loss

Focal Loss for Dense Object Detection， Lin etc, ICCV 2017 Best student paper

Page 20: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Two-Stage detector: FPN/Mask R-CNN

• FPN Structure

• ROIAlign

Mask R-CNN， He etc, ICCV 2017 Best paper

Page 21: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

What is next for object detection?

• The pipeline seems to be mature

• There still exists a large gap between existing state-of-arts and product requirements

• The devil is in the detail

Page 22: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Outline

• Introduction to Object Detection• Modern Object detectors

• One Stage detector vs Two-stage detector• Challenges

• Backbone• Head• Pretraining• Scale• Batch Size• Crowd• NAS• Fine-Grained

• Conclusion

Page 23: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Challenges Overview

• Backbone• Head• Pretraining• Scale• Batch Size• Crowd• NAS• Fine-grained

Backbone HeadPostprocess

NMS

Page 24: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Challenges - Backbone

• Backbone network is designed for classification task but not for localization task

• Receptive Field vs Spatial resolution

• Only f1-f5 is pretrained but randomly initializing f6 and f7 (if applicable)

Page 25: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Backbone - DetNet

• DetNet: A Backbone network for Object Detection, Li etc, 2018, https://arxiv.org/pdf/1804.06215.pdf

https://arxiv.org/pdf/1804.06215.pdf
Page 26: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Backbone - DetNet

Page 27: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Backbone - DetNet

Page 28: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Backbone - DetNet

Page 29: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Backbone - DetNet

Page 30: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Backbone - DetNet

Page 31: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Challenges - Head

• Speed is significantly improved for the two-stage detector

• RCNN - > Fast RCNN -> Faster RCNN - > RFCN

• How to obtain efficient speed as one stage detector like YOLO, SSD?

• Small Backbone

• Light Head

Page 32: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Head – Light head RCNN

• Light-Head R-CNN: In Defense of Two-Stage Object Detector, 2017, https://arxiv.org/pdf/1711.07264.pdf

Code: https://github.com/zengarden/light_head_rcnn

https://arxiv.org/pdf/1711.07264.pdf
Page 33: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Head – Light head RCNN

• Backbone

• L: Resnet101

• S: Xception145

• Thin Feature map

• L:C_{mid} = 256

• S: C_{mid} =64

• C_{out} = 10 * 7 * 7

• R-CNN subnet

• A fc layer is connected to the PS ROI pool/Align

Page 34: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Head – Light head RCNN

Page 35: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Head – Light head RCNN

Page 36: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Head – Light head RCNN

• Mobile Version

• ThunderNet: Towards Real-time Generic Object Detection, Qin etc, Arxiv2019

• https://arxiv.org/abs/1903.11752

https://arxiv.org/abs/1903.11752
Page 37: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Pretraining – Objects365

• ImageNet pretraining is usually employed for backbone training

• Training from Scratch

• Scratch Det claims GN/BN is important

• Rethinking ImageNet Pretraining validates that training time is important

Page 38: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Pretraining – Objects365

• Objects365 Dataset

Page 39: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Pretraining – Objects365

• Pretraining with Objects365 vs ImageNet vs from Sctratch

Page 40: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Pretraining – Objects365

• Pretraining on Backbone or Pretraining on both backbone and head

Page 41: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Pretraining – Objects365

• Results on VOC Detection & VOC Segmentation

Page 42: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Pretraining – Objects365

• Summary

• Pretraining is important to reduce the training time

• Pretraining with a large dataset is beneficial for the performance

Page 43: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Challenges - Scale

• Scale variations is extremely large for object detection

Page 44: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Challenges - Scale

• Scale variations is extremely large for object detection

• Previous works

• Divide and Conquer: SSD, DSSD, RON, FPN, …

• Limited Scale variation

• Scale Normalization for Image Pyramids, Singh etc, CVPR2018

• Slow inference speed

• How to address extremely large scale variation without compromising inference speed?

Page 45: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Scale - SFace

• SFace: An Efficient Network for Face Detection in Large Scale Variations, 2018, http://cn.arxiv.org/pdf/1804.06559.pdf

• Anchor-based:

• Good localization for the scales which are covered by anchors

• Difficult to address all the scale ranges of faces

• Anchor-free:

• Able to cover various face scales

• Not good for the localization ability

http://cn.arxiv.org/pdf/1804.06559.pdf
Page 46: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Scale - SFace

Page 47: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Scale - SFace

Page 48: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Scale - SFace

Page 49: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Scale - SFace

• Summary:

• Integrate anchor-based and anchor-free for the scale issue

• A new benchmark for face detection with large scale variations: 4K Face

Page 50: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Challenges - Batchsize

• Small mini-batchsize for general object detection

• 2 for R-CNN, Faster RCNN

• 16 for RetinaNet, Mask RCNN

• Problem with small mini-batchsize

• Long training time

• Insufficient BN statistics

• Inbalanced pos/neg ratio

Page 51: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Batchsize – MegDet

• MegDet: A Large Mini-Batch Object Detector, CVPR2018, https://arxiv.org/pdf/1711.07240.pdf

https://arxiv.org/pdf/1711.07240.pdf
Page 52: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Batchsize – MegDet

• Techniques

• Learning rate warmup

• Cross-GPU Batch Normalization

Page 53: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Challenges - Crowd

• NMS is a post-processing step to eliminate multiple responses on one object instance

• Reasonable for mild crowdness like COCO and VOC

• Will Fail in the case when the objects are in a crowd

Page 54: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Challenges - Crowd

• A few works have been devoted to this topic

• Softnms, Bodla etc, ICCV 2017, http://www.cs.umd.edu/~bharat/snms.pdf

• Relation Networks, Hu etc, CVPR 2018, https://arxiv.org/pdf/1711.11575.pdf

• Lacking a good benchmark for evaluation in the literature

http://www.cs.umd.edu/~bharat/snms.pdf
https://arxiv.org/pdf/1711.11575.pdf
Page 55: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Crowd - CrowdHuman

• CrowdHuman: A Benchmark for Detecting Human in a Crowd, 2018, https://arxiv.org/pdf/1805.00123.pdf, http://www.crowdhuman.org/

• A benchmark with Head, Visible Human, Full body bounding-box

• Generalization ability for other head/pedestrian datasets

• Crowdness

https://arxiv.org/pdf/1805.00123.pdf
http://www.crowdhuman.org/
Page 56: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Crowd - CrowdHuman

Page 57: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Crowd-CrowdHuman

Page 58: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Crowd-CrowdHuman

• Generalization

• Head

• Pedestrian

• COCO

Page 59: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Conclusion

• The task of object detection is still far from solved

• Details are important to further improve the performance

• Backbone

• Head

• Pretraining

• Scale

• Batchsize

• Crowd

• The improvement of object detection will be a significantly boost for the computer vision industry

Page 60: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

广告部分

• Megvii Detection 知乎专栏

Email: [email protected]

Page 61: Beyond RetinaNet and Mask R-CNN › Presentation › ICME2019_tutorial...2019/07/08  · Schedule of Tutorial •Lecture 1: Beyond RetinaNet and Mask R-CNN (Gang Yu) •Lecture 2:

Mask R-CNN - Facebook Research · PDF fileMask R-CNN is conceptually simple: Faster R-CNN has two outputs for each candidate object, a class label and a ... (RPN), proposes candidate

Mask R-CNN - Facebook Research · PDF fileMask R-CNN is conceptually simple: Faster R-CNN has two outputs for each candidate object, a class label and a ... (RPN), proposes candidate

Mask R-CNN and OBIA Fusion Improves the Segmentation of ...

Mask R-CNN and OBIA Fusion Improves the Segmentation of ...

Deep Learning for Image Instance Segmentation ----Mask R-CNN · object instance segmentation. •Mask R-CNN, extends Faster R-CNN by adding a branch for predicting segmentation masks

Deep Learning for Image Instance Segmentation ----Mask R-CNN · object instance segmentation. •Mask R-CNN, extends Faster R-CNN by adding a branch for predicting segmentation masks

Face mask detection using YOLOv3 and faster R-CNN models ...

Face mask detection using YOLOv3 and faster R-CNN models ...

Machine Learning for Video Tracking of Passengers and ...Convolutional Pose Machine [1] Mask R-CNN [2] [1] Shih-En, Ramakrishna, Kanade, Sheikh. "Convolutional pose machines." IEEE

Machine Learning for Video Tracking of Passengers and ...Convolutional Pose Machine [1] Mask R-CNN [2] [1] Shih-En, Ramakrishna, Kanade, Sheikh. "Convolutional pose machines." IEEE

CNN-PS: CNN-based Photometric Stereo for General Non ...openaccess.thecvf.com/...CNN-PS_CNN-based_Photometric_ECCV_2… · CNN-PS: CNN-based PhotometricStereo for General Non-Convex

CNN-PS: CNN-based Photometric Stereo for General Non ...openaccess.thecvf.com/...CNN-PS_CNN-based_Photometric_ECCV_2… · CNN-PS: CNN-based PhotometricStereo for General Non-Convex

2sRanking-CNN: A 2-stage ranking-CNN for …2.2.1 1st-stage Ranking-CNN Before explaining 1st-stage ranking-CNN, let’s briefly explain how ranking-CNN works. Ranking-CNN was proposed

2sRanking-CNN: A 2-stage ranking-CNN for …2.2.1 1st-stage Ranking-CNN Before explaining 1st-stage ranking-CNN, let’s briefly explain how ranking-CNN works. Ranking-CNN was proposed

Detección y tracking de objetos mediante la Mask R-CNN

Detección y tracking de objetos mediante la Mask R-CNN

CNN | by CNN Staff

CNN | by CNN Staff

Estimating 2D Multi-Hand Poses From Single Depth Images€¦ · Keywords: Multi-Hand Pose Estimation · Pictorial Structure · Mask R-CNN 1 Introduction Accurate hand pose estimation

Estimating 2D Multi-Hand Poses From Single Depth Images€¦ · Keywords: Multi-Hand Pose Estimation · Pictorial Structure · Mask R-CNN 1 Introduction Accurate hand pose estimation

Mask R-CNN - openaccess.thecvf.comopenaccess.thecvf.com/content_ICCV_2017/papers/He_Mask_R-CNN_… · Mask R-CNN Kaiming He Georgia Gkioxari Piotr Doll´ar Ross Girshick Facebook

Mask R-CNN - openaccess.thecvf.comopenaccess.thecvf.com/content_ICCV_2017/papers/He_Mask_R-CNN_… · Mask R-CNN Kaiming He Georgia Gkioxari Piotr Doll´ar Ross Girshick Facebook

arXiv:1809.08545v3 [cs.CV] 16 Apr 2019 · 3.1. Bounding Box Parameterization Based on a two-stage object detector Faster R-CNN or Mask R-CNN [42,17] shown in Figure3, we propose to

arXiv:1809.08545v3 [cs.CV] 16 Apr 2019 · 3.1. Bounding Box Parameterization Based on a two-stage object detector Faster R-CNN or Mask R-CNN [42,17] shown in Figure3, we propose to

Optimizing the Trade-off between Single-Stage and Two-Stage … · Faster R-CNN (Region-based Convolutional Neural Networks) or Mask R-CNN, that (i) use a Region Proposal Network

Optimizing the Trade-off between Single-Stage and Two-Stage … · Faster R-CNN (Region-based Convolutional Neural Networks) or Mask R-CNN, that (i) use a Region Proposal Network

CNN Airport Media Kiti.cdn.turner.com/cnn/CNN/Programs/airport.network/cnn...Programming includes breaking news, business and financial reports, sports updates, weather, lifestyle

CNN Airport Media Kiti.cdn.turner.com/cnn/CNN/Programs/airport.network/cnn...Programming includes breaking news, business and financial reports, sports updates, weather, lifestyle

Meta R-CNN : Towards General Solver for Instance-level Low ... · inference in Faster /Mask R-CNNs. Each RoI feature refers to a single object or background, so Faster /Mask R-CNN

Meta R-CNN : Towards General Solver for Instance-level Low ... · inference in Faster /Mask R-CNNs. Each RoI feature refers to a single object or background, so Faster /Mask R-CNN

Mask r-cnn · 2018-02-09 · Overview of Mask R-CNN •Goal: to create a framework for Instance segmentation •Builds on top of Faster R-CNN by adding a parallel branch •For each

Mask r-cnn · 2018-02-09 · Overview of Mask R-CNN •Goal: to create a framework for Instance segmentation •Builds on top of Faster R-CNN by adding a parallel branch •For each

Mask R-CNN for Moving Shadow Detection and Segmentation

Mask R-CNN for Moving Shadow Detection and Segmentation

Feasibility Study: Digitalisation of Collaborative Human ......CPU K80 1070 ti 2080 ti Mask R-CNN (1024x1024) DensePose (512x512) DeepLab (512x512) OpenPose (512x512) 4. Data fusion

Feasibility Study: Digitalisation of Collaborative Human ......CPU K80 1070 ti 2080 ti Mask R-CNN (1024x1024) DensePose (512x512) DeepLab (512x512) OpenPose (512x512) 4. Data fusion

learning model pit mine, Vietnam using the Mask R-CNN deep ...

learning model pit mine, Vietnam using the Mask R-CNN deep ...

See table 2.b. in Mask RCNN paper - Ablation section · Mask R-CNN Kaiming He Georgia Gkioxari Piotr Doll´ar Ross Girshick Facebook AI Research (FAIR) Abstract We present a conceptually

See table 2.b. in Mask RCNN paper - Ablation section · Mask R-CNN Kaiming He Georgia Gkioxari Piotr Doll´ar Ross Girshick Facebook AI Research (FAIR) Abstract We present a conceptually