The whole network starts! Super detailed sparser CNN actual combat
This article comes from the 3D visual developer community and is written by Wang Hao 1. Introduction At present, there are two main methods in the field of target detection. The first category is the deny detector, which has been widely used since the non Deep era, such as DPM, YOLO, RetinaNet UTF-8...
Posted by flash_78 on Tue, 26 Oct 2021 12:31:20 +0530
[learn the compiler from scratch] 12. Learning notes of MLIR Toy Tutorials 1
This note is a summary of learning MLIR Tutorials. Criticism and correction are welcome. Chapter1: Toy language and AST MLIR provides a toy language to explain the definition and implementation process of MLIR. Toy language is a Tensor based language. We can use it to define functions, perform UTF-8...
Posted by reli4nt on Tue, 26 Oct 2021 21:09:24 +0530
Deep learning pytoch -- comparison of self defined automatic derivation function + pytoch and TensorFlow in core summary 2
Deep learning pytoch (VII) -- core summary 2: self defined automatic derivation function + comparison of networks built by pytoch and TensorFlow 1, Define a new automatic derivation function At the bottom, each original automatic derivation operation is actually two functions running on Tensor.UTF-8...
Posted by datona on Fri, 29 Oct 2021 02:03:03 +0530
Code note: variable prototype encoder: one shot learning with prototype images
Official code https://github.com/mibastro/VPE 0. Preparation data set The address of the data set is given in the README.md file. Download it and unzip it. You can also download Baidu cloud Link: https://pan.baidu.com/s/1-4E-ixSuhpQ9r-LX3DAZGA Extraction code: f1xw Then modify the path in the UTF-8...
Posted by PHP Man on Fri, 29 Oct 2021 09:53:56 +0530
Color planet image generation 4: transpose convolution + interpolation scaling + convolution shrinkage (pytorch version)
Previous episode: Color planet image generation 3: code improvement (pytorch version) Based on the previous code, more modifications have been made to improve the generation effect. The chessboard effect of transpose convolution is further optimized, and some work is also done in other aspectsUTF-8...
Posted by otuatail on Fri, 29 Oct 2021 12:45:19 +0530
NLP : RNN / Attention based seq2seq
This article is the reading notes of advanced deep learning: natural language processing seq2seq model seq2seq means "(from) sequence to sequence", that is, one sequence data is converted to another Encoder decoder model The seq2seq model is also known as the Encoder Decoder model. As the name UTF-8...
Posted by 938660 on Fri, 29 Oct 2021 17:42:55 +0530
The whole process record of C + + pytorch model
As recorded above, the process is such a process: Configure libtorch -- > How does the python training model work in C + +_ Jihai Guyu CSDN blogpytorch model transformationWrite C + + caller Here we will record the model transformation and C + + calling program. There's not much to say about thUTF-8...
Posted by JamesWebster on Fri, 29 Oct 2021 22:50:18 +0530
Deep learning neural network introduction case detailed analysis - Iris case
Neural network design process Case: iris classification There are three categories of iris: Three species: dog tail weed small abdominal muscle A neural network is built to classify iris Collect the characteristic values of flowers: four kinds Calyx length Calyx width Petal length Petal width AUTF-8...
Posted by gillms1 on Mon, 01 Nov 2021 00:11:11 +0530
[mmdetection] RetinaNet analysis take RetinaNet as an example to analyze the anchor generation, matching, encoding and decoding strategies in target detection
1. RetinaNet one-stage detector Innovation: RetinaNet Network + Focal loss to solve the imbalance between positive and negative samplesStructure: backbone + FPN + head (bbox & class) 2. Configuration file retinanet_r50_fpn.py is parsed as follows: backbone to configure model = dict( type='RetiUTF-8...
Posted by wtg21.org on Mon, 01 Nov 2021 20:03:41 +0530
Deep learning -- TensorFlow (project) verification code generation and identification (multi task learning)
catalogue Basic theory 1, Generate verification code data set 1. Generate verification code training set 1-0. Judge whether the folder is empty 1-1. Create a character set (numbers, uppercase and lowercase letters) 1-2. Randomly generate verification codes (1000, length 4) 2. Generate verificatUTF-8...
Posted by lucianoes on Tue, 02 Nov 2021 22:35:37 +0530