• http://www.thinkface.cn/

Tool

• network definition visualization

Environment Setup

• GPU深度学习环境搭建之软件篇

Frameworks

Mini-Caffe

• https://github.com/luoyetx/mini-caffe
• 【C++版】采用mini-caffe框架实现的人脸对齐源码下载，mini-caffe框架介绍

ncnn

• Ubuntu16.04---腾讯NCNN框架入门到应用

Mobile

• F-CNN：用FPGA框架训练卷积神经网络

SeetaFace

• https://github.com/seetaface/SeetaFaceEngine
• 【开源】SEETAFACE开源人脸识别引擎！｜XTECHER要闻
• Jetson ARM SeetaFace编译

Performance Metrics (e.g. FLOPS)

• FLOPS与GOPS：各平台及神经网络算力算量调研
• 玩转 FLOPS – ARM 如何测量 GPU 计算性能
• how to calculate a net's FLOPs in CNN
• https://github.com/Lyken17/pytorch-OpCounter
• https://github.com/1adrianb/pytorch-estimate-flops
• How to understand / calculate FLOPs of the neural network model?

Accuracy Metrics

• 【YOLO学习】召回率（Recall），精确率（Precision），平均正确率（Average_precision(AP) ），交除并（Intersection-over-Union（IoU））
• Evaluation Metrics for Machine Learning - Accuracy, Precision, Recall, and F1 Defined
• Intersection over Union (IoU) for object detection

smooth L1 loss

• 损失函数：L1 loss, L2 loss, smooth L1 loss
• 请问faster rcnn和ssd 中为什么用smooth l1 loss，和l2有什么区别？

NMS

• NMS——非极大值抑制

Normalization

• 一文弄懂Batch Norm / Layer Norm / Instance Norm / Group Norm 归一化方法

• Instance Normalisation vs Batch normalisation

• 模型优化之Layer Normalization

• 【基础算法】六问透彻理解BN(Batch Normalization）

• 从几何视角来理解模型参数的初始化策略

• BN究竟起了什么作用？一个闭门造车的分析

• 浅谈Transformer的初始化、参数化与标准化

LayerNorm

• 为什么Pre Norm的效果不如Post Norm？

DeepNorm

• [细读经典]DeepNet - 1000 layer Transformers

OP

RNN

• https://en.wikipedia.org/wiki/Recurrent_neural_network

• https://stanford.edu/~shervine/teaching/cs-230/cheatsheet-recurrent-neural-networks

• Understanding RNN and LSTM

• Illustrated Guide to LSTM’s and GRU’s: A step by step explanation

• RNN 的梯度消失问题

• Optimizing Recurrent Neural Networks in cuDNN 5

LSTM

• 谷歌大脑科学家亲解 LSTM：一个关于“遗忘”与“记忆”的故事
• LSTM模型与前向反向传播算法
• https://developer.nvidia.com/discover/lstm

ROI Pooling / ROI Align

• Understanding Region of Interest (RoI Pooling)

• Understanding Region of Interest - Part 2 (RoI Align)

• 图解RoI Align & RoI Warp

• 图解 RoIPooling、RoIAlign 和 RoIWarp

• onnx实现

  • https://github.com/microsoft/onnxruntime/blob/master/onnxruntime/core/providers/cpu/object_detection/roialign.cc

• tensorflow实现

  • https://github.com/tensorpack/tensorpack/blob/6d5ba6a970710eaaa14b89d24aace179eb8ee1af/examples/FasterRCNN/model.py#L301
  • https://github.com/longcw/RoIAlign.pytorch/blob/master/roi_align/roi_align.py

• Region of interest pooling explained

• Faster RCNN原理篇（二）——RoIPooling和RoIAlign的学习和理解

• 【摘记】ROI Pooling and ROI Align

RROI Pooling

• https://github.com/mjq11302010044/RRPN/blob/master/caffe-fast-rcnn/src/caffe/layers/rotate_roi_pooling_layer.cpp
• https://github.com/mjq11302010044/RRPN/blob/master/caffe-fast-rcnn/src/caffe/layers/rotate_roi_pooling_layer.cu

RROI Align

• https://github.com/mjq11302010044/RRPN/blob/master/caffe-fast-rcnn/src/caffe/layers/rotate_roi_align_layer.cu

• Image segmentation with Mask R-CNN

• https://github.com/facebookresearch/Detectron

• https://github.com/ShaoqingRen/faster_rcnn

Resize

• https://github.com/onnx/onnx/blob/master/docs/Operators.md#Resize
• 是什么引起了各个框架 Resize 操作的结果不同？——来自 ONNX 的标准化尝试
• https://github.com/onnx/onnx/pull/3026
• https://github.com/onnx/onnx/blob/master/onnx/backend/test/case/node/resize.py

Softmax

• 人脸识别中Softmax-based Loss的演化史

ScatterND

• http://www.xavierdupre.fr/app/mlprodict/helpsphinx/onnxops/onnx__ScatterND.html
• [TOPI][CUDA] “scatter_nd”’ has a very poor performance on CUDA backend (>1000x slower than hand-written cuda code)
• [技术干货] 【昇腾众智】RetinaFace_ResNet50模型的910训练+310推理

OP Fusion

• 关于fusion（算子合并）的一些总结
• Deep Dive into Kernel Fusion: Accelerating Inference in Llama V2

Networks

Feedforward neural network

• Feedforward neural network
• What is a Feed Forward Neural Network?
• Deep Learning (Part 1) - Feedforward neural networks (FNN)

mtcnn

• 人脸检测——MTCNN
• https://github.com/Seanlinx/mtcnn
• https://github.com/kpzhang93/MTCNN_face_detection_alignment
• MTCNN人脸及特征点检测---代码应用详解（基于ncnn架构）

FDDB

• https://github.com/hualitlc/MTCNN-on-FDDB-Dataset
• https://github.com/itlab-vision/FDDB/tree/master/fddb_evaluate

Object Detection

Faster-RCNN

• Object Detection and Classification using R-CNNs
• 一文读懂Faster RCNN

Image Segmentation

Mask-RCNN

• 令人拍案称奇的Mask RCNN

• A Brief History of CNNs in Image Segmentation: From R-CNN to Mask R-CNN

• ROI-pooling and ROI-align

• Region of interest pooling explained

• How does ROTAlign work in Mask-RCNN?

• How to calculate feature map after applying ROI Align, as explained in Mask RCNN Paper?

• https://github.com/matterport/Mask_RCNN/issues/128

• https://github.com/yangxue0827/R-DFPN_FPN_Tensorflow/blob/master/libs/rotation/rbbox_overlaps_kernel.cu

• Position Detection and Direction Prediction for Arbitrary-Oriented Ships via Multitask Rotation Region Convolutional Neural Network

PointRend

• Ross、何恺明等人提出PointRend：渲染思路做图像分割，显著提升Mask R-CNN性能
• PointRend: Image Segmentation as Rendering

Yolo v3

• https://github.com/tanluren/yolov3-channel-and-layer-pruning

• How to Implement a YOLO (v3) Object Detector from Scratch in PyTorch: Part 1

• How to implement a YOLO (v3) object detector from scratch in PyTorch: Part 2

• How to implement a YOLO (v3) object detector from scratch in PyTorch: Part 3

• YOLOV3剪枝源码阅读---模型部署加速

• YOLOV3算法详解

• 万字长文，用代码的思想讲解Yolo3算法实现原理，Visdrone数据集和自己制作数据集两种方式在Pytorch训练Yolo模型

DCN

• 论文阅读：DCNv1：Deformable Convolutional Networks

Darknet

• darknet yolo v2 训练自己的数据