[cs231n] Lecture 11 | Detection and Segmentation

Lecture 11 | Detection and Segmentation

---

• Semantic Segmentation
• Object Detection
• Instance Segmentation 

---

Semantic Segmentation

Problem: Very inefficient! Not reusing shared features between overlapping patches

Problem: convolutions at original image resolution will be very expensive ...

downsampling and upsampling

Downsampling: Pooling, strided convolution

---

In-Network upsampling: “Unpooling”

In-Network upsampling: “Max Unpooling”

Learnable Upsampling: Transpose Convolution

Transpose(Deconvolution)

1D Example

Matrix Multiplication

---

Semantic Segmentation Idea: Fully Convolutional

Multi-view 3D Reconstruction

2D Object Detection

---

Classification + Localization

 

Object Detection as Regression?

Each image needs a different number of outputs!

Object Detection as Classification: Sliding Window

---

 Region Proposals / Selective Search: R-CNN

 

Problems

---

Fast R-CNN

Fast R-CNN: RoI Pooling

 

R-CNN vs SPP vs Fast R-CNN

 

Faster R-CNN

---

Mask R-CNN

---

Detection without Proposals: YOLO / SSD

---

Object Detection + Captioning = Dense Captioning

VisualGenome

Scene Graph Generation

3D Object Detection

Simplified Camera Model

Scale & Distance Ambiguity