https://github.com/tensorflow/models/tree/master/research TensorFlow Research Models
This folder contains machine learning models implemented by researchers in TensorFlow. The models are maintained by their respective authors. To propose a model for inclusion, please submit a pull request.
Currently, the models are compatible with TensorFlow 1.0 or later. If you are running TensorFlow 0.12 or earlier, please upgrade your installation.
adversarial_crypto: protecting communications with adversarial neural cryptography.
adversarial_text: semi-supervised sequence learning with adversarial training.
attention_ocr: a model for real-world image text extraction.
audioset: Models and supporting code for use with AudioSet.
autoencoder: various autoencoders.
brain_coder: Program synthesis with reinforcement learning.
cognitive_mapping_and_planning: implementation of a spatial memory based mapping and planning architecture for visual navigation.
compression: compressing and decompressing images using a pre-trained Residual GRU network.
deeplab: deep labelling for semantic image segmentation.
delf: deep local features for image matching and retrieval.
differential_privacy: privacy-preserving student models from multiple teachers.
domain_adaptation: domain separation networks.
gan: generative adversarial networks.
im2txt: image-to-text neural network for image captioning.
inception: deep convolutional networks for computer vision.
learning_to_remember_rare_events: a large-scale life-long memory module for use in deep learning.
lexnet_nc: a distributed model for noun compound relationship classification.
lfads: sequential variational autoencoder for analyzing neuroscience data.
lm_1b: language modeling on the one billion word benchmark.
maskgan: text generation with GANs.
namignizer: recognize and generate names.
neural_gpu: highly parallel neural computer.
neural_programmer: neural network augmented with logic and mathematic operations.
next_frame_prediction: probabilistic future frame synthesis via cross convolutional networks.
object_detection: localizing and identifying multiple objects in a single image.
pcl_rl: code for several reinforcement learning algorithms, including Path Consistency Learning.
ptn: perspective transformer nets for 3D object reconstruction.
qa_kg: module networks for question answering on knowledge graphs.
real_nvp: density estimation using real-valued non-volume preserving (real NVP) transformations.
rebar: low-variance, unbiased gradient estimates for discrete latent variable models.
resnet: deep and wide residual networks.
skip_thoughts: recurrent neural network sentence-to-vector encoder.
slim: image classification models in TF-Slim.
street: identify the name of a street (in France) from an image using a Deep RNN.
swivel: the Swivel algorithm for generating word embeddings.
syntaxnet: neural models of natural language syntax.
tcn: Self-supervised representation learning from multi-view video.
textsum: sequence-to-sequence with attention model for text summarization.
transformer: spatial transformer network, which allows the spatial manipulation of data within the network.
video_prediction: predicting future video frames with neural advection.
https://github.com/tensorflow/models/tree/master/research/slim
https://blog.csdn.net/qq_36356761/article/details/79521694
https://github.com/dmlc/gluon-cv
https://github.com/pytorch/examples
https://github.com/pytorch/vision (this is torchvision source code)
(1)Image classification (MNIST) using Convnets
(2)Word level Language Modeling using LSTM RNNs
(3)Training Imagenet Classifiers with Residual Networks
(4)Generative Adversarial Networks (DCGAN)
(5)Variational Auto-Encoders
(6)Superresolution using an efficient sub-pixel convolutional neural network
(7)Hogwild training of shared ConvNets across multiple processes on MNIST
(8)Training a CartPole to balance in OpenAI Gym with actor-critic
(9)Natural Language Inference (SNLI) with GloVe vectors, LSTMs, and torchtext
(10)Time sequence prediction - use an LSTM to learn Sine waves
(11)Implement the Neural Style Transfer algorithm on images
(12)Several examples illustrating the C++ Frontend
(1)Perceptual Losses for Real-Time Style Transfer and Super-Resolution(fast-neural-style)
https://www.jianshu.com/p/b728752a70e9
lua:https://github.com/jcjohnson/fast-neural-style
tensorflow:https://github.com/hzy46/fast-neural-style-tensorflow
pytorch:https://github.com/abhiskk/fast-neural-style
chainer:https://github.com/yusuketomoto/chainer-fast-neuralstyle
(2)Texture Networks: Feed-forward Synthesis of Textures and Stylized Images(texture_nets)
https://www.jianshu.com/p/1187049ae1ad
tensorflow:https://github.com/tgyg-jegli/tf_texture_net
lua:https://github.com/DmitryUlyanov/texture_nets
(3)Instance Normalization: The Missing Ingredient for Fast Stylization(IN)
https://www.jianshu.com/p/d77b6273b990
lua:https://github.com/DmitryUlyanov/texture_nets
(4)Fast Neural Style Transfer with Arbitrary Style using AdaIN Layer(AdaIN)
lua:https://github.com/xunhuang1995/AdaIN-style
pytorch:https://github.com/naoto0804/pytorch-AdaIN
tensorflow:https://github.com/elleryqueenhomels/arbitrary_style_transfer
(5)Controlling Perceptual Factors in Neural Style Transfer
lua:https://github.com/leongatys/NeuralImageSynthesis
(6)Deep Photo Style Transfer
https://blog.csdn.net/cicibabe/article/details/70868746
lua:https://github.com/luanfujun/deep-photo-styletransfer
tensorflow:https://github.com/NVIDIA/FastPhotoStyle
tensorflow:https://github.com/LouieYang/deep-photo-styletransfer-tf
pytorch:https://github.com/NVIDIA/FastPhotoStyle
(7)Unpaired Image-to-Image Translation Using Cycle-Consistent Adversarial Networks(cyclegan)
torch:https://github.com/junyanz/CycleGAN
pytorch:https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix
(1)DeblurGAN: Blind Motion Deblurring Using Conditional Adversarial Networks
https://arxiv.org/pdf/1711.07064.pdf
keras:https://github.com/RaphaelMeudec/deblur-gan
(1)STACL: Simultaneous Translation with Integrated Anticipation and Controllable Latency
https://arxiv.org/pdf/1810.08398.pdf
(1)Look at Boundary: A Boundary-Aware Face Alignment Algorithm
https://github.com/wywu/LAB
(1)Mobile Net v1(MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications)
caffe:https://github.com/shicai/MobileNet-Caffe
(2)Mobile Net v2(MobileNetV2: Inverted Residuals and Linear Bottlenecks)
caffe:https://github.com/shicai/MobileNet-Caffe
(3)Resnet(Deep Residual Learning for Image Recognition)
pytorch:https://github.com/Cadene/pretrained-models.pytorch
(4)VGG(Very Deep Convolutional Networks for Large-Scale Image Recognition)
(5)Google Net
https://blog.csdn.net/u011974639/article/details/76460849#googlenet
Inception V1:Going deeper with convolutions
Inception V2:Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
Inception V3:Rethinking the Inception Architecture for Computer Vision
Inception V4:Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning
https://github.com/hoya012/deep_learning_object_detection
my blog:https://blog.csdn.net/u013066730/article/details/82460392
https://github.com/tensorflow/models/tree/master/research/object_detection
https://paperswithcode.com/sota/object-detection-on-coco(This is a good web.)
(1)maskrcnn https://arxiv.org/abs/1703.06870
mx:https://github.com/TuSimple/mx-maskrcnn
tf:https://github.com/CharlesShang/FastMaskRCNN
keras+tf:https://github.com/matterport/Mask_RCNN
pytorch:https://github.com/facebookresearch/maskrcnn-benchmark
caffe2:https://github.com/facebookresearch/Detectron
(2)FCIS https://arxiv.org/abs/1611.07709
https://github.com/msracver/FCIS
(3)SSD http://arxiv.org/abs/1512.02325
caffe:https://github.com/weiliu89/caffe/tree/ssd
(4)M2Det https://qijiezhao.github.io/imgs/m2det.pdf
pytorch:https://github.com/qijiezhao/M2Det
(5)efficientdet
keras:https://github.com/xuannianz/EfficientDet
(6)yolov3
keras:https://github.com/OlafenwaMoses/ImageAI#detection
(7)efficientdet
tensorflow:https://github.com/google/automl/tree/master/efficientdet
https://github.com/Cadene/pretrained-models.pytorch
https://github.com/keras-team/keras-applications
https://github.com/qubvel/classification_models
(1)PCANet https://arxiv.org/pdf/1404.3606.pdf
chainer:https://github.com/IshitaTakeshi/PCANet
scalar c++:https://github.com/Ldpe2G/PCANet
(2)CBAM&BAM:https://github.com/Jongchan/attention-module
(3)efficientnet
keras:https://github.com/qubvel/efficientnet
https://blog.playment.io/semantic-segmentation-models-autonomous-vehicles/
https://github.com/mrgloom/awesome-semantic-segmentation
https://mp.weixin.qq.com/s/w7pYxm52QbcFPRBe12iMdA and https://arxiv.org/abs/2001.05566
好用：https://github.com/qubvel/segmentation_models.pytorch and https://github.com/qubvel/segmentation_models
(1)Linknet:https://arxiv.org/abs/1707.03718
https://codeac29.github.io/projects/linknet/
lua:https://github.com/mjiansun/LinkNet
pytorch:https://github.com/e-lab/pytorch-linknet
https://github.com/willard-yuan/awesome-cbir-papers
tensorflow:https://github.com/hwalsuklee/tensorflow-generative-model-collections
pytorch:https://github.com/znxlwm/pytorch-generative-model-collections
article:https://github.com/zhangqianhui/AdversarialNetsPapers
(1)stargan
pytorch:https://github.com/yunjey/StarGAN
(2)cyclegan
pytorch:https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix
(3)wgan
pytorch:https://github.com/caogang/wgan-gp
(4)staingan
article&code:https://xtarx.github.io/StainGAN/
https://github.com/xinario/awesome-gan-for-medical-imaging
RDSNet:https://arxiv.org/abs/1912.05070
https://github.com/wangsr126/RDSNet
YOLACT:https://arxiv.org/abs/1904.02689
https://github.com/dbolya/yolact
YOLACT++:https://arxiv.org/abs/1912.06218
https://github.com/dbolya/yolact
CenterMask:https://arxiv.org/abs/1911.06667
https://github.com/youngwanLEE/CenterMask
maskrcnn:https://arxiv.org/abs/1703.06870
https://github.com/matterport/Mask_RCNN
https://github.com/DeepRegNet/DeepReg
https://github.com/ZhiningLiu1998/awesome-imbalanced-learning
以下参考：https://blog.csdn.net/nature553863/article/details/81083955
模型压缩算法能够有效降低参数冗余，从而减少存储占用、通信带宽和计算复杂度，有助于深度学习的应用部署，具体可划分为如下几种方法：
(1)线性或非线性量化：1/2bits, int8 和 fp16等
模型量化是指权重或激活输出可以被聚类到一些离散、低精度（reduced precision）的数值点上，通常依赖于特定算法库或硬件平台的支持：
- 二值化网络：XNORnet [https://arxiv.org/abs/1603.05279, Github: https://github.com/ayush29feb/Sketch-A-XNORNet, Github: https://github.com/jiecaoyu/XNOR-Net-PyTorch], ABCnet with Multiple Binary Bases [https://arxiv.org/abs/1711.11294, Github: https://github.com/layog/Accurate-Binary-Convolution-Network], Bin-net with High-Order Residual Quantization [https://arxiv.org/abs/1708.08687], Bi-Real Net [https://arxiv.org/abs/1808.00278, Github: https://github.com/liuzechun/Bi-Real-net]；
- 三值化网络：Ternary weight networks [https://arxiv.org/abs/1605.04711], Trained Ternary Quantization [https://arxiv.org/abs/1612.01064, Github: https://github.com/czhu95/ternarynet]；
- W1-A8 或 W2-A8量化： Learning Symmetric Quantization [http://phwl.org/papers/syq_cvpr18.pdf, Github: https://github.com/julianfaraone/SYQ]；
- INT8量化：TensorFlow-lite [https://arxiv.org/abs/1712.05877], TensorRT [http://on-demand.gputechconf.com/gtc/2017/presentation/s7310-8-bit-inference-with-tensorrt.pdf], Quantization Interval Learning [https://blog.csdn.net/nature553863/article/details/96857133]；
- INT4量化：NVIDIA Iterative Online Calibration [https://blog.csdn.net/nature553863/article/details/104080434], LSQ [https://blog.csdn.net/nature553863/article/details/104275477];
- 其他（非线性）：Intel INQ [https://arxiv.org/abs/1702.03044], log-net, CNNPack [https://papers.nips.cc/paper/6390-cnnpack-packing-convolutional-neural-networks-in-the-frequency-domain] 等；
- Post-training量化策略：针对预训练模型，通过适当调整kernel参数分布、或补偿量化误差，可有效提升量化效果；
- 关于量化的比较系统性的论述，参考论文：Quantizing deep convolutional networks for efficient inference: A whitepaper；
(2)结构或非结构剪枝：deep compression, channel pruning 和 network slimming等；
非结构剪枝：通常是连接级、细粒度的剪枝方法，精度相对较高，但依赖于特定算法库或硬件平台的支持，如Deep Compression [https://arxiv.org/abs/1510.00149], Sparse-Winograd [https://arxiv.org/abs/1802.06367, https://ai.intel.com/winograd-2/, Github: https://github.com/xingyul/Sparse-Winograd-CNN] 算法等；
结构剪枝：是filter级或layer级、粗粒度的剪枝方法，精度相对较低，但剪枝策略更为有效，不需要特定算法库或硬件平台的支持，能够直接在成熟深度学习框架上运行:
- 如局部方式的、通过layer by layer方式的、最小化输出FM重建误差的Channel Pruning [https://arxiv.org/abs/1707.06168, Github: https://github.com/yihui-he/channel-pruning], ThiNet [https://arxiv.org/abs/1707.06342], Discrimination-aware Channel Pruning [https://arxiv.org/abs/1810.11809, Github: https://github.com/Tencent/PocketFlow]；
- 全局方式的、通过训练期间对BN层Gamma系数施加L1正则约束的Network Slimming [https://arxiv.org/abs/1708.06519, Github: https://github.com/foolwood/pytorch-slimming]；
- 全局方式的、按Taylor准则对Filter作重要性排序的Neuron Pruning [https://arxiv.org/abs/1611.06440, Github: https://github.com/jacobgil/pytorch-pruning]；
- 全局方式的、可动态重新更新pruned filters参数的剪枝方法 [http://xuanyidong.com/publication/ijcai-2018-sfp/]；
- 基于GAN思想的GAL方法 [https://blog.csdn.net/nature553863/article/details/97631176]，可裁剪包括Channel, Branch或Block等在内的异质结构；
- 借助Geometric Median确定卷积滤波器冗余性的剪枝策略 [https://blog.csdn.net/nature553863/article/details/97760040]；
- 基于Reinforcement Learning (RL)，实现每一层剪枝率的连续、精细控制，并可结合资源约束完成自动模型压缩 [https://github.com/mit-han-lab/amc];
(3)网络结构搜索 (NAS: Network Architecture Search)：DARTS, DetNAS、NAS-FCOS、Proxyless NAS和NetAdapt等；
(4)其他：权重矩阵的低秩分解，知识蒸馏与网络结构简化（squeeze-net, mobile-net, shuffle-net）等；
知识蒸馏介绍：https://blog.csdn.net/u013066730/article/details/111573882
百度的知识蒸馏SSLD：https://github.com/PaddlePaddle/PaddleClas/blob/master/docs/zh_CN/advanced_tutorials/distillation/distillation.md
知识蒸馏论文代码整理：https://github.com/dkozlov/awesome-knowledge-distillation
论文：Exploring knowledge distillation of DNNs for efficient hardware solutions, https://github.com/peterliht/knowledge-distillation-pytorch
https://tefuirnever.blog.csdn.net/article/details/90719309
https://zhuanlan.zhihu.com/p/31664818
https://www.cnblogs.com/carsonzhu/p/10860594.html
EDSR:https://github.com/sanghyun-son/EDSR-PyTorch
noise2noise:https://github.com/NVlabs/noise2noise
Pytorch2caffe:https://github.com/xxradon/PytorchToCaffe
chainer:https://github.com/chainer
pytorch:https://github.com/pytorch
pydicom:https://github.com/pydicom
mxnet:https://github.com/dmlc
baidu:https://github.com/baidu-research
tencent：https://github.com/Tencent
tencent Ai Lab：https://ai.tencent.com/ailab/index.html
microsoft:https://github.com/Microsoft
facebook:https://github.com/facebookresearch
google:https://github.com/google
opencv:https://github.com/opencv
deepinsight:https://github.com/deepinsight
cmusatyalab:https://github.com/cmusatyalab
nvidia:https://github.com/NVIDIA
Purdue University:https://github.com/e-lab
cvpr,iccv:http://openaccess.thecvf.com/menu.py
eccv:https://dblp.uni-trier.de/db/conf/eccv/index.html
https://github.com/stlndm/linke https://github.com/ChinaArJun/Tencent-NOW
1）Stanford:https://www.coursera.org/learn/machine-learning
code:https://github.com/yoyoyohamapi/mit-ml