Rico's Nerd Cluster

「离开世界之前 一切都是过程」

Deep Learning - Performance Metrics

mean Average Precision (mAP), Precision, Recall, ROC Curve, F1 Score

Terminology Area Under Curve Area Under Curve = AUC. True Positives, False Positives, True Negatives, False Negatives Suppose a class has 10 students, 5 boys and 5 girls. You use a machine to f...

Deep Learning - Image Segmentation

Encoder-Decoder, Fully-Convolutional Networks (FCN), U-Net

Pre-Requitstes Encoder-Decoder framework Autoencoders (or encoders) and autodecoders (or decoders) were introduced in the late 1980. An autoencoder compresses input data into smaller dimensions, ...

Deep Learning - Object Detection Notes Part 2

R-CNN

Region Based CNN (R-CNN, Girshick et al. CVPR 2014) Zhihu Regional Proposal is the core of R-CNN. It first uses a segmentation algorithm to find regions with objects, then use these regions as “r...

Deep Learning - Object Detection Notes Part 1

Convolution Implementation of Sliding Window, OverFeat

Introduction Image classification is given an image, output a class lable of the image Image classification and Object localization is Object Detection. For example, if there are 4 classes, [Pede...

Deep Learning - Hands-On ResNet Transfer Learning For CIFAR-10 Dataset

Data Normalization, Conv Net Training

ResNet-50 Transfer Learning COMPLETE CODE can be found here Data Loading Please see this blogpost for data loading Model Definition PyTorch Built-In Model 1 2 3 4 model = models.resnet50(weig...

Deep Learning - CNN Applications

TensorFlow Keras Sequential and Functional Models

TF Keras Sequential API TF Keras Sequential API can be used to build simple feedforward, sequential models. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 ...

Deep Learning - Classic CNN Models

LeNet-5, AlexNet, VGG-16, ResNet-50, One-by-One Convolution, Inception Network

LeNet-5 (65K params) The LeNet-5 architecture (LeCun, 1998) and is still very widely used. 5 is simply the version number. LeNet-5 has 7 layers. Input layer are 32x32 grayscale images. MNIST i...

Deep Learning - CNN Basics

Filters, Padding, Convolution and Its Back Propagation, Receptive Field

Filters Filters (aka kernels): “Pattern Detectors”. Each filter is a small matrix, which you can drag along an image and multiply pixel values with (convolution). They can detect edges, corners, a...

Deep Learning - Start Easy, Things I Learned From Training Small Neural Nets

Basic Torch Network With Some Notes on Syntax

Introduction To gain some insights into how hyper parameters impacts training, I created a simple neural network using PyTorch to learn 2D input data. Specifically, I’m interested in exploring the...

Deep Learning - PyTorch Basics

Neural Network Model Components, Common Operations

Data Type Conversions Common Data Types torch.arange(start, stop, step) can take either float or int values torch.range(start, stop, step) is deprecated because its signature is dif...