This paper will summarize the harvesting training camp depth study carried out on the Baidu platform is based on AI Studio
theory
Some problems faced by image recognition processing:
1. semantic gap (Semantic Gap) phenomenon:
The gap between the bottom and the visual characteristics of the image level semantic concept
teacher simply is : similar visual characteristics, different semantic concept;
dissimilar visual characteristics of the same semantic concept;
The image recognition basic frame
It represents a measuring space characterized by the feature space, matching the feature space by the features of the category space.
Performance characteristics expressed in the traditional method: design features; performance on the depth of learning: learning feature.
3. The image recognition processing flow Early
(1) Feature extraction : global feature extraction, i.e., with global visual characteristics of the underlying image representing statistics
global feature if: color, texture feature, shape and other characteristics
but also the earlier problems of image processing technology: global feature lost image details
(2) indexing techniques : binary hash
reducing characteristics of the storage space, and wherein the similarity calculation complexity
(3) Relevance feedback : The intended to infer that the user can observe the behavior
(4) reordering : more similar to the initial alignment is rearranged watches, is selected by the visual information.
4. The mid-frame image recognition
(1) Feature extraction :
local features (Local Feature): image block (Patch) vector
wherein the detector Feature Detector (detection center position of the image block); characterized descriptor Feature Descriptor (block described visual content)
(2 ) vectorization :
(3) Indexing :
(4) Post-treatment :
to enhance the high consistency results matching score; reduced low consistency score matching results.
The concept of artificial intelligence:
here, the teacher's words to explain my feelings very deeply , is to remove human sensibility, leaving the human reason, so that its maximum benefit of this is artificial intelligence.
Machine learning process:
Depth study of the concept:
Established model to mimic the human brain, neural networks use this function to solve the problem of machine learning, so called multi-layers of depth.
Based on image recognition depth study of
image classification: For a given image, the predicted image-level tag
The first day of practice Harvest
The first day of practice for the task: making real-time visualization of the epidemic
This task for me, difficult, involving reptiles, and python foundation is very general, not been exposed to reptiles, but the teacher has other similar examples, part reptile has been given, the final completion of the enthusiastic group of friends to help pointing job.
In my understanding this task is divided into two parts, as follows: reptile section; another part is: use Baidu open source data visualization tools Echarts, which contains a lot of sophisticated graphic design, good interaction, to find a pie chart in which the API pie configuration, a series of global configuration and configuration.
The end result effect is as follows:
This figure reflects the cumulative number of people diagnosed nationwide in 2020 March 31
Summary: Baidu familiar with the use of AI Studio platform to facilitate the compilation environment to build on the notebook, an insight into the powerful open source features
The above pictures are from Baidu AI Studio platform, you can be interested to learn something Ha!
https://aistudio.baidu.com/aistudio/index
Codeword is not easy, your support is the greatest driving force of my work, attention to the next point like it
