Scheduling curves of 12 learning rate schedulers +OneCycleLR built in paddlepaddle 23 paddle

At present, based on this base class, 12 strategies have been implemented in the pad, namely: cosineanealingdelay, exponentialdelay, inversetimedecay, lambdadecay, linearwarmup, multistepdelay, naturalexpecay, noamdecay, piecewisedelay, polynomialdecay, reduceonplateau, stepdelay. In addition, bloggers sorted out OneCycleLR according to the onl ...

Posted by rjlowe on Fri, 08 Apr 2022 14:36:20 +0300

Use your own data set to train the use of C3D and tensorboard

Modify C3D As soon as I ran, I was excited to send a document. I can only run. The recognition accuracy has not been studied. It's basically modification mypath.py dataset.py train.py inference.py ucf_labels.txt these files ucf_labels.txt First of all, I modified it based on ucf101, so I modified ucf_labels.txt the label in this file has bee ...

Posted by phpnewbie1979 on Fri, 08 Apr 2022 12:14:51 +0300

Study notes of AlexNet, VGG and GoogleNet

1. Foundation work for deep learning: AlexNet AlexNet won the ILSVRC 2012 (ImageNet Large Scale Visual Recognition Challenge) competition in 2012, with top1 error: 36.7% and top5 error:15.4% SIFT+FVS, the second traditional machine learning method at that time, top 5 error: 26.2%, and traditional machine learning has been difficult to conti ...

Posted by daedalus__ on Fri, 08 Apr 2022 10:34:26 +0300

Cluster deep learning training practice notes -- high performance distributed multi task data reading

Cluster deep learning training practice notes -- high performance distributed multi task data reading FesianXu 20220406 at Baidu Search Team preface In [1,2], the author previously introduced some data reading experience in distributed training in large-scale clusters. This paper will continue the previous content and continue to s ...

Posted by PHPeter on Thu, 07 Apr 2022 16:26:28 +0300

WSL2-Ubuntu20.04 configure deep learning environment (CUDA, CUDNN, pytoch)

WSL2-Ubuntu20.04 configure deep learning environment (CUDA, CUDNN, pytoch) 1, Foreword Now there are different opinions on the online tutorials, which are not particularly complete. This post aims to record the pits encountered in installing wsl2. 2, Install WSL2 WSL2 is essentially different from WSL1. Due to the replacement of the whole a ...

Posted by billy2shoe on Wed, 06 Apr 2022 08:44:10 +0300

tensorflow2.3 realize PPLCNet -- a lightweight backbone network with faster speed and higher accuracy

1. Preface PPLCNet is a lightweight and high-performance network PP lcnet designed by Baidu team in combination with the end-to-side reasoning characteristics of Intel CPU. The proposed scheme achieves better delay accuracy balance than ShuffleNetV2, MobileNetV2, MobileNetV3 and GhostNet in image classification tasks. This paper proposes a li ...

Posted by lakilevi on Wed, 06 Apr 2022 07:36:56 +0300

China Mobile integration - resumption of the first OneCity Programming Competition

China Mobile integration - resumption of the first OneCity Programming Competition Official account: ChallengeHub Competition link: https://js.dclab.run/v2/cmptDetail.html?id=457 Code link of this article: https://github.com/yanqiangmiffy/One-City-Challenge/tree/master/code 1. Competition background The theme of this OneCity programming co ...

Posted by Arsench on Tue, 05 Apr 2022 22:18:36 +0300

Transfer learning demonstration of computer vision

This paper describes how to use transfer learning to train a convolutional neural network for image classification task. For more on migration learning, see cs231n notes. About these notes: In fact, people usually don't train a whole convolutional neural network from scratch (starting with the random initialization weight), because there is u ...

Posted by happypete on Mon, 04 Apr 2022 09:06:07 +0300

Machine learning notes - use CNN and LSTM to generate text descriptions for images

1, Task description Seeing an image, your brain can easily distinguish what the image is about, but can the computer distinguish the content represented by the image? With the progress of deep learning technology, the availability of large data sets and computer capabilities, we can build models that can generate descriptions for images. The ...

Posted by kayess2004 on Mon, 04 Apr 2022 04:30:36 +0300

Detailed explanation of YOLO v3 source code

0. Summary Recently, I studied yolo3, read many blogs and understood some theoretical knowledge, but it was still a little hard to learn. After reading the source code, I had a further understanding. Here, I will not repeat the network code. The network code is easy to understand. The following will give a detailed explanation and analysis of t ...

Posted by samdennis on Mon, 04 Apr 2022 01:07:15 +0300