Programmer Think

Alibaba cloud OSS object storage upload pictures and intelligent image recognition garbage recognition I believe that the majority of novice programmers, like me, want to write an intelligent image recognition, but they can't find it in Baidu or don't know what it means when they check. Next, I will contribute my own process and code of writin ...

This blog is only used to record the learning process, avoid forgetting and facilitate future review Convolutional neural network is good at image processing. It is inspired by human visual nervous system. CNN has two characteristics: 1. It can effectively reduce the dimension of pictures with large amount of data into small amount of ...

Training a Classifier We learned how to define the neural network, calculate the loss and update the network weight What about data? Usually, when you have to deal with images, text, audio or video, you can use a standard python package that can load data into a numpy array, and then convert the array to torch* Tensor Image: pilot, opencvAu ...

Series index: Chapter 1 Python Basics (1) Python Libraries - numpy, matplotlib Here is an example to understand the use of two important Python libraries. import numpy as np import matplotlib.pyplot as plt x = np.arange(0, 6, 0.1) y1 = np.sin(x) y2 = np.cos(x) plt.plot(x, y1, label="sin") plt.plot(x, y2, linestyle="--", label ...

[source code analysis] NVIDIA HugeCTR, GPU version parameter server - (4) 0x00 summary In this series, we introduce HugeCTR, an industry-oriented recommendation system training framework, which is optimized for large-scale CTR models with model parallel embedding and data parallel intensive networks. This paper mainly introduces the firs ...

1, Common skills in transfer learning: fine tuning 1.1 concept Take the weights trained on the big data set as the initialization weight of the specific task (small data set), and retrain the network (modify the full connection layer output as needed); The training methods can be: 1. Fine tune all layers; 2. Fix the weights of the front l ...

overview For those detectors running on GPU platform, their backbone network may be VGG, ResNet, ResNeXt or DenseNet. For those detectors running on the CPU platform, their detectors may be SqueezeNet, MobileNet and shufflenet. The most representative two-stage target detector R-CNN series include fast R-CNN, fast R-CNN, R-FCN [9], Libra ...

preface: from torch.utils.data import Dataset from torch.utils.data import DataLoader import torchvision.transforms as transforms 1. Dataset in torch utils. data 2. It is required for loading text or image data sets, custom data sets, special formats or official data sets. DataLoader in 3,torch.utils.data uses this package for both text a ...

1. Ideas There are some similarities between BM3D and NLM algorithms. NLM has written before: go to They all use the features of other regions of the image and the current block features to fuse into denoised image blocks. The main differences are as follows:         1. Search only within a fixed radi ...

Reprinted from AI Studio Title item link https://aistudio.baidu.com/aistudio/projectdetail/3401596 PP minilm, a fast, small and accurate Chinese characteristic pre training model, has been released! Transformer pre training model has achieved remarkable results in NLP tasks, but its huge model parameters and slow inference speed limit its wi ...