Week5-Faster R-CNN.pptx

Faster R-CNN
By; Deep Learning Team

History(?) of R-CNN
• Rich feature hierarchies for accurate object detection and semantic segmentation(2013)
• Fast R-CNN(2015)
• Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks(2015)
• Mask R-CNN(2017)

Is Faster R-CNN
Really Fast?
• Generally R-FCN and SSD
models are faster on
average while Faster R-
CNN models are more
accurate
• Faster R-CNN models can
be faster if we limit the
number of regions
proposed

Region Proposals – Selective Search
• Bottom-up segmentation, merging regions at multiple scales
Convert
regions to
boxes

R-CNN Training
• Pre-train a ConvNet(AlexNet) for ImageNet classification dataset
• Fine-tune for object detection(softmax + log loss)
• Cache feature vectors to disk
• Train post hoc linear SVMs(hinge loss)
• Train post hoc linear bounding-box regressors(squared loss)
“Post hoc” means the parameters are learned after
the ConvNet is fixed

Bounding-Box Regression
specifies the pixel coordinates of the center of proposal Pi’s
bounding box together with Pi’s width and height in pixels
means the ground-truth bounding box

Problems of R-CNN
• Slow at test-time: need to run full forward path of CNN for
each region proposal
 13s/image on a GPU(K40)
 53s/image on a CPU
• SVM and regressors are post-hoc: CNN features not updated
in response to SVMs and regressors
• Complex multistage training pipeline (84 hours using K40
GPU)
 Fine-tune network with softmax classifier(log loss)
 Train post-hoc linear SVMs(hinge loss)
 Train post-hoc bounding-box regressions(squared loss)

Fast R-CNN
• Fix most of what’s wrong with R-CNN and SPP-net
• Train the detector in a single stage, end-to-end
 No caching features to disk
 No post hoc training steps
• Train all layers of the network

RoI Pooling
RoI in Conv feature map : 21x14  3x2 max pooling with stride(3, 2)  output : 7x7
RoI in Conv feature map : 35x42  5x6 max pooling with stride(5, 6)  output : 7x7
VGG-16

Training & Testing
1. Takes an input and a set of bounding boxes
2. Generate convolutional feature maps
3. For each bbox, get a fixed-length feature vector from RoI
pooling layer
4. Outputs have two information
 K+1 class labels
 Bounding box locations
• Loss function

R-CNN vs SPP-net vs Fast R-CNN
Runtime dominated by
region proposals!

Problems of Fast R-CNN
• Out-of-network region proposals are the test-time
computational bottleneck
• Is it fast enough??

Faster R-CNN(RPN + Fast R-CNN)
• Insert a Region Proposal
Network (RPN) after the last
convolutional layer  using GPU!
• RPN trained to produce region
proposals directly; no need for
external region proposals
• After RPN, use RoI Pooling and
an upstream classifier and bbox
regressor just like Fast R-CNN

Training Goal : Share Features

3 x 3
RPN
• Slide a small window on the
feature map
• Build a small network for
 Classifying object or not-object
 Regressing bbox locations
• Position of the sliding window
provides localization information
with reference to the image
• Box regression provides finer
localization information with
reference to this sliding window
ZF : 256-d, VGG : 512-d

RPN
• Use k anchor boxes at each
location
• Anchors are translation
invariant: use the same ones at
every location
• Regression gives offsets from
anchor boxes
• Classification gives the
probability that each
(regressed) anchor shows an
object

3 x 3
RPN(Fully Convolutional Network)
• Intermediate Layer – 256(or 512)
3x3 filter, stride 1, padding 1
• Cls layer – 18(9x2) 1x1 filter, stride
1, padding 0
• Reg layer – 36(9x4) 1x1 filter, stride
1, padding 0
ZF : 256-d, VGG : 512-d

Anchors as references
• Anchors: pre-defined reference boxes
• Multi-scale/size anchors:
 Multiple anchors are used at each position:
3 scale(128x128, 256x256, 512x512) and 3 aspect rations(2:1, 1:1, 1:2) yield 9
anchors
 Each anchor has its own prediction function
 Single-scale features, multi-scale predictions

Positive/Negative Samples
• An anchor is labeled as positive if
 The anchor is the one with highest IoU overlap with a ground-truth
box
 The anchor has an IoU overlap with a ground-truth box higher than
0.7
• Negative labels are assigned to anchors with IoU lower than
0.3 for all ground-truth boxes
• 50%/50% ratio of positive/negative anchors in a minibatch

Is It Enough?
• RoI Pooling has some quantization operations
• These quantizations introduce misalignments between the RoI
and the extracted features
• While this may not impact classification, it can make a
negative effect on predicting bbox

Mask R-CNN
• Mask R-CNN extends Faster R-CNN by adding a branch for
predicting segmentation masks on each Region of Interest
(RoI), in parallel with the existing branch for classification and
bounding box regression

Loss Function, Mask Branch
• The mask branch has a K x m x m - dimensional output for
each RoI, which encodes K binary masks of resolution m × m,
one for each of the K classes.
• Applying per-pixel sigmoid
• For an RoI associated with ground-truth class k, Lmask is only
defined on the k-th mask

RoI Align
• RoI Align don’t use quantization of the RoI boundaries
• Bilinear interpolation is used for computing the exact values
of the input features

Week5-Faster R-CNN.pptx

More Related Content

Similar to Week5-Faster R-CNN.pptx (20)

More from fahmi324663 (16)

Recently uploaded (20)

Week5-Faster R-CNN.pptx