De Jun Huang

Learning Day 71: Image captioning

Image captioning

• Vision → Language

Learning Day 70: 3D U-Net with 3D convolution layers, V-Net, DenseNet, FC-DenseNet

Learning Day 69: Image segmentation for biomedical applications — U-Net

U-Net

Learning Day 68: Semantic segmentation 2 — DeepLab, atrous/dilated convolution

Background

Learning Day 67: Semantic segmentation 1 — FCN; Deconvolution

Image segmentation

Learning Day 66: Object detection 5 — YOLO v1, v2 and v3

YOLO v1 (You Only Look Once)

Learning Day 65: Object detection 4 — R-FCN

Past models

R-CNN

Learning Day 64: Object detection 3 — Fast R-CNN and Faster R-CNN

Fast R-CNN

Learning Day 63: Object detection 2— SPP-Net

SPP-Net

• Improve on the drawbacks of R-CNN: slow…

Learning Day 62: Object detection — R-CNN

Object detection

• Detect an object + identify what…

Learning Day 61:Deep learning based image retrieval

Deep learning based Binary Hash Codes

• Make use of CNN and transfer learning to add additional latent layer
• The concept of latent layer is similar to auto-encoder…

Learning Day 60: Image search via traditional CV methods 2

Continue from Day 59, other traditional…

Learning Day 59: Image search via traditional CV methods 1

Image search

• Provide an image and search for similar images
• Basically a image similarity problem
• There are traditional CV and deep learning…

Learning Day 58: Revise on classic CNN models and new take-aways

Revise on CNN models in a different…

Learning Day 57/Practical 5: Loss function — CrossEntropyLoss vs BCELoss in Pytorch; Softmax vs sigmoid; Loss calculation

CrossEntropyLoss vs BCELoss

1. Difference in purpose

Learning Day 56/Practical 4: Retrieving raw scores from CNN models

Retrieving raw scores from output layers

Objectives

• To look at the raw scores and determine if the prediction is confident

Learning Day 55: back propagation in CNN — pooling and conv layers

Building the knowledge on top of…

Learning Day 54/Practical 3: Extracting feature maps from other conv layers

Continue from Day 53 to extract feature maps from other layers

• It is very similar as getting the first feature map from conv layer 1

Learning Day 53/Practical 2: Extracting kernels and feature map from conv layer

Know-hows from past…

Learning Day 52: Back propagation — a simple example for hand calculation

A simple 2-layer NN for hand…

Learning Day 51: Back propagation details

Back propagation for a simple NN

Learning Day 50: Revise on NN and CNN in another course and new take-aways

New take-aways for NN

Learning Day 49: Take a break from reading, start practicing — building my own dataset in Pytorch

Learning Day 49/Practical 1: Building my own dataset in Pytorch from HDF5

Learning Day 48: Computer vision before deep learning 2 — Face detection with Haar cascade and pedestrian detection…