Recommended 論文紹介: Fast R-CNN&Faster R-CNN
SSD: Single Shot MultiBox Detector (ECCV2016)
[DL輪読会] Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields
[DL輪読会]VNect: Real-time 3D Human Pose Estimation with a Single RGB Camera
20130626 kawasaki.rb NKT77
20130626 kawasaki.rb NKT77
20180427 arXivtimes 勉強会: Cascade R-CNN: Delving into High Quality Object Det...
[DL輪読会]YOLO9000: Better, Faster, Stronger
[DL輪読会]Focal Loss for Dense Object Detection
物体検知（Meta Study Group 発表資料）
ArtTrack: Articulated Multi-Person Tracking in the Wild : CV勉強会関東
Variational Kalman Filter
Shusaku Taniguchi Bachelor Thesis
プログラム説明 kgPhotonMapping v0-1-0
輪読発表資料: Efficient Virtual Shadow Maps for Many Lights
[DL輪読会]Encoder-Decoder with Atrous Separable Convolution for Semantic Image S...
[DL輪読会]Parallel WaveNet: Fast High-Fidelity Speech Synthesis
[DL輪読会]Differentiable Mapping Networks: Learning Structured Map Representatio...
Rabbit challenge dnn_day3-4
How to Schedule Machine Learning Workloads Nicely In Kubernetes #CNDT2020 / C...
レトリバ勉強会資料：深層学習による自然言語処理２章
【2015.07】(1/2)cvpaper.challenge@CVPR2015
【2015.05】cvpaper.challenge@CVPR2015
More Related Content 論文紹介: Fast R-CNN&Faster R-CNN
SSD: Single Shot MultiBox Detector (ECCV2016)
[DL輪読会] Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields
[DL輪読会]VNect: Real-time 3D Human Pose Estimation with a Single RGB Camera
20130626 kawasaki.rb NKT77
20130626 kawasaki.rb NKT77
What's hot (20) 20180427 arXivtimes 勉強会: Cascade R-CNN: Delving into High Quality Object Det...
[DL輪読会]YOLO9000: Better, Faster, Stronger
[DL輪読会]Focal Loss for Dense Object Detection
物体検知（Meta Study Group 発表資料）
ArtTrack: Articulated Multi-Person Tracking in the Wild : CV勉強会関東
Variational Kalman Filter
Shusaku Taniguchi Bachelor Thesis
プログラム説明 kgPhotonMapping v0-1-0
輪読発表資料: Efficient Virtual Shadow Maps for Many Lights
[DL輪読会]Encoder-Decoder with Atrous Separable Convolution for Semantic Image S...
[DL輪読会]Parallel WaveNet: Fast High-Fidelity Speech Synthesis
[DL輪読会]Differentiable Mapping Networks: Learning Structured Map Representatio...
Rabbit challenge dnn_day3-4
How to Schedule Machine Learning Workloads Nicely In Kubernetes #CNDT2020 / C...
レトリバ勉強会資料：深層学習による自然言語処理２章
Similar to AGA_CVPR2017 (20) 【2015.07】(1/2)cvpaper.challenge@CVPR2015
【2015.05】cvpaper.challenge@CVPR2015
Learning Spatial Common Sense with Geometry-Aware Recurrent Networks
cvpaper.challenge@CVPR2015(Attribute)
【DLゼミ】XFeat: Accelerated Features for Lightweight Image Matching
CVPR 2019 report (30 papers)
Fast, Accurate Detection of 100,000 Object Classes on a Single Machine
（文献紹介）深層学習による動被写体ロバストなカメラの動き推定
これからのコンピュータビジョン技術 - cvpaper.challenge in PRMU Grand Challenge 2016 (PRMU研究会 2...
Domain Adaptive Faster R-CNN for Object Detection in the Wild 論文紹介
KantoCV/Selective Search for Object Recognition
点群SegmentationのためのTransformerサーベイ
PFI成果発表会2014発表資料 Where Do You Look?
CVPR2016読み会 "Inside-Outside Net: Detecting Objects in Context with Skip Pooli...
3. Data Augmentation
3
Cifar-10, 9層のConvNet [1]
Method error
Without DA 9.08%
With DA 7.25%
With Large DA 4.41%
[1] JT Springenberg, Striving for Simplicity: The All Convolutional Net
学習データを人工的に増やす
flipping random cropping
6. Tables with depth
in the range of 1-2[m]
6
𝐱 = 𝜙 𝐱, 𝑡
目的：𝜙の学習
𝑠. 𝑡. 𝛾 𝐱 = 𝑡
Input: 画像特徴𝐱
output: 画像特徴 𝐱
s.t. アトリビュート𝛾 𝐱 = 𝑡
8. Attribute Guided Augmentation
8
Attribute regressor 𝛾 𝐱 : 𝒳 → ℝ+
Feature regressor 𝜙𝑖
𝑘
𝐱 : 𝒳 → 𝒳
𝑖: 入力特徴のアトリビュート（区間）
𝑘: 出力特徴のアトリビュート
𝒳: feature space (an object)
入力区間数 × 出力ターゲット数 × アトリビュート数
の𝜙𝑖
𝑘
を学習する
15. 15
Depth [m] Pose [deg]Median
absolute error
同一クラスで学習 vs クラスを無視して学習
Depth
0.2m, 7.5m
Pose
0°, 180°
データが少ない
(lamp, door)と厳
しい
20. Object-based one-shot scene recognition
20
物体検出ネットワークの特徴ベクトルから
シーン認識を行う
AggregateFast RCNN
0.2
−0.8
⋮
0.4
Images from A. Gupta, From 3D Scene Geometry to Human Workspace
SVM
![Data Augmentation
3
Cifar-10, 9層のConvNet [1]
Method error
Without DA 9.08%
With DA 7.25%
With Large DA 4.41%
[1] JT Springenberg, Striving for Simplicity: The All Convolutional Net
学習データを人工的に増やす
flipping random cropping](https://image.slidesharecdn.com/cvpr2017agaupload-170618015918/85/AGA_CVPR2017-3-320.jpg)
![Tables with depth
in the range of 1-2[m]
6
𝐱 = 𝜙 𝐱, 𝑡
目的：𝜙の学習
𝑠. 𝑡. 𝛾 𝐱 = 𝑡
Input: 画像特徴𝐱
output: 画像特徴 𝐱
s.t. アトリビュート𝛾 𝐱 = 𝑡](https://image.slidesharecdn.com/cvpr2017agaupload-170618015918/85/AGA_CVPR2017-6-320.jpg)
![15
Depth [m] Pose [deg]Median
absolute error
同一クラスで学習 vs クラスを無視して学習
Depth
0.2m, 7.5m
Pose
0°, 180°
データが少ない
(lamp, door)と厳
しい](https://image.slidesharecdn.com/cvpr2017agaupload-170618015918/85/AGA_CVPR2017-15-320.jpg)
