计算机视觉学习 1

1.What problem does the cloud solve when considering real-world computer vision problems?

• In the traditional hardware model architecture, if higher processing power or more storage space is needed, users usually choose the vertical expansion mode, that is, to purchase more advanced and powerful servers to achieve, but the vertical expansion capacity of the system is always limited. With the development of computer vision, the application scale is getting larger and larger, and the number of users is increasing, it is difficult, if not impossible, for physical machines to meet the demand. Also for large-scale applications, it is more expensive to build a system using a vertical expansion model. Cloud computing uses standardized, low-cost hardware, and then scales horizontally through software to build a large and stable computing platform. This platform not only outperforms traditional mainframes, but also costs a fraction of the cost of traditional mainframes. In the current difficult development of computer processors, cloud computing technology will reduce the need for servers and storage devices, and will reduce the need for PCs.

• In addition, cloud computing and platforms can 1) gradually break down the data barriers existing in different industries; 2) reduce the investment in the construction of multimedia and visualization information systems and improve the utilization of visualization assets.

2. How could colab notebooks and/or Jupyter Books be used to exchange ideas or build research portfolios?

• Google Colab gives AI enthusiasts and developers a free GPU on which they can Tensorflow, PyTorch and other deep learning frameworks. colab real-time notebooks facilitate developers in data sharing, and files can be shared with other developers through links. When the file is finished in the notebook, you can either save it to Google Drive or upload it to a GitHub repository. Once you\'ve committed the file to the GitHub repository, you can open it from your GitHub account using the shortcut link at the top of the file. If you want to share the GitHub file, you can click the "share" button in the top right corner. Other people can view, comment, and edit it.

 

3. What are some key differences between Biological and Machine Vision?

• Machine vision is a branch of artificial intelligence that is rapidly evolving. Simply put, machine vision is the use of machines instead of the human eye to make measurements and judgments. Machine vision system is through the image ingestion device will be ingested target into image signal, transmitted to the special image processing system, to get the morphological information of the target, according to the pixel distribution and brightness, color and other information, into digital signals; image system to these signals for various operations to extract the characteristics of the target, and then according to the results of the discrimination to control the action of the equipment on site.

• Vision is a physiological term. Light acts on the visual organ to excite its receptor cells, and its information is processed by the visual nervous system to produce vision. Through vision, humans and animals perceive the size, light and darkness, color, movement, and various information important to the survival of the organism, at least 80% of external information obtained by vision, vision is one of the important senses of humans and animals.

• Historical experience shows that the study of the visual system of primates, especially humans, has contributed to the study of computer vision. At present, numerous computer vision algorithms designed to simulate the biological vision system are widely used in practical applications. However,

1. biological vision has nominally guided computer vision, but progress has actually been slower than the latter;

2. the development of computer vision has mostly come from intuition and experience, and the relationship with biological vision is only at the conceptual level.

• The Professor Songchun Zhu is opposed to the idea that the current research on computer vision requires first understanding how biological vision systems work. His arguments are twofold:

1. the discovery of biological vision systems often lags behind the proposal of specific algorithms in the field of computer vision;

2. computer vision algorithms have strong practical applications and perform far better than biological vision algorithms in specific vision tasks.

• Compared to biological vision, the disadvantages of machine vision also exist:

1. existing machine vision systems are relatively single-function, and cannot process multiple information at the same time as intelligent creatures;

2. machine vision systems, all of which require specialized programming and system design by professional technicians, are still very rare, but biological vision is something that every intelligent creature is born with;

3. existing image recognition uses machine learning methods for image understanding, generally for specific image categories and fixed application scenarios, do not yet have the real sense of visual perception and visual understanding of intelligent creatures. Moreover, intelligent creatures rely on three-dimensional senses for external space and objects, and current image machine learning algorithms, which are generally based on two-dimensional image data, have inherent defects and weaknesses in terms of perception and understanding of space.