|
所在平台: Udemy |
课程主页: https://www.udemy.com/course/convolutional-neural-networks-for-image-classification/
课程评论：没有评论
本课程“卷积神经网络用于图像分类”将带你深入学习并实践卷积神经网络（CNN）在图像分类任务中的应用。 课程初期，你将从零开始，使用纯 NumPy 库和循环，在灰度图像上实现卷积、池化操作以及它们的组合，并利用不同滤波器可视化物体边缘检测和实时跟踪。 随后，课程将指导你如何组合图像、创建自定义数据集用于分类任务，并将数据集保存为二进制文件。其中，你将学习如何将现有的交通标志数据集转换为适合分类任务的格式，并保存。接着，将进行数据预处理，并将处理后的数据集保存。 核心部分，你将构建 CNN 模型，选择合适的层数和参数以实现高精度分类。完成模型设计和数据集准备后，你将训练 CNN 模型，并在新图像上进行测试，实现实时摄像头图像分类，并可视化训练过程中滤波器的变化。 最后，你将通过实践测试来巩固所学知识。作为额外福利，课程还将教授如何通过图像旋转、投影和亮度调整等方法生成百万级额外图像，扩充现有数据集。 本课程旨在提升你应用 CNN 解决实际图像分类问题的硬技能。每节课都设有 SMART（特定、可衡量、可实现、结果导向、时限性）目标，让你能够清晰地追踪学习进度并立即看到实践成果。
In this practical course, you'll design, train and test your own Convolutional Neural Network (CNN) for the tasks of Image Classification.By the end of the course, you'll be able to build your own applications for Image Classification.At the beginning, you'll implement convolution, pooling and combination of these two operations to grayscale images by the help of different filters, pure Numpy library and 'for' loops. We will also implement convolution in Real Time by camera to detect objects edges and to track objects movement.After that, you'll assemble images together, compose custom dataset for classification tasks and save created dataset into a binary file.Next, you'll convert existing dataset of Traffic Signs into needed format for classification tasks and save it into a binary file.Then, you'll apply preprocessing techniques before training, produce and save processed datasets into separate binary files.At the next step, you'll construct CNN models for classification tasks, select needed number of layers for accurate classification and adjust other parameters.When the models are designed and datasets are ready, you'll train constructed CNNs, test trained models on completely new images, classify images in Real Time by camera and visualize training process of filters from randomly initialized to finally trained.At the final step, you'll pass Practice Test according to the all learned material during the course.As a bonus part, you'll generate up to 1 million additional images and extend prepared dataset by new images via image rotation, image projection and brightness changing. The main goal of the course is to develop and improve your hard skills in order to apply them for real problems of Image Classification based on Convolutional Neural Networks.Every lecture of the course has SMART objectives. It means, that you can track your progress and witness practical results within the visible time frame, right after the end of the lecture.S - specific (the lecture has specific objectives)M - measurable (results are reasonable and can be quantified)A - attainable (the lecture has clear steps to achieve the objectives)R - result-oriented (results can be obtained by the end of the lecture)T - time-oriented (results can be obtained within the visible time frame)
