Unlock instant, AI-driven research and patent intelligence for your innovation.
Controllable automobile image synthesis method based on causal flow model
What is Al technical title?
Al technical title is built by PatSnap Al team. It summarizes the technical point description of the patent document.
An image synthesis, automotive technology, applied in the field of image processing, can solve problems such as the connection between conditional information and images, the difficulty of accurately measuring the underlying distribution of image conditions, and multiple targets.
Active Publication Date: 2020-12-18
CHONGQING UNIV
View PDF5 Cites 2 Cited by
Summary
Abstract
Description
Claims
Application Information
AI Technical Summary
This helps you quickly interpret patents by identifying the three key elements:
Problems solved by technology
Method used
Benefits of technology
Problems solved by technology
[0004]
At present, there is a deep learning architecture based on flow model and conditional encoder that can overcome the above shortcomings. The reversibility and accurate log-likelihood of the image space and latent space mapping of the flow model have great potential in image synthesis. In the encoder, the coded input attribute annotation is used as a supervisory condition as a controllable factor in the generated image, which can preserve controllable information. Such a model must perform a bijective mapping between the distribution of the image and the latent vector, that is, its latent dimension must match the visible dimensionality, but there is no way to connect the conditional information with the image into the full model
Therefore, a straightforward idea is to add class-dependent regularization in the optimization objective. However, when encountering complex situations, model training often fails. The reason for this phenomenon is the image condition on the latent space. The underlying distribution is difficult to measure accurately, and there are multiple targets
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more
Image
Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
Click on the blue label to locate the original text in one second.
Reading with bidirectional positioning of images and text.
Smart Image
Examples
Experimental program
Comparison scheme
Effect test
Embodiment 1
[0127] A controllable car image synthesis method based on a causal flow model, comprising the following steps:
[0128] 1) Get the original car image data and write it into the car image dataset D. Preprocess the car image data set D to get the car image data set D′=[D 1 ,D 2 ,...,D X ]. X is the total number of car image samples. D. X Represents a car image sample.
[0129] The original car image data is Stanford car image data. The Stanford car images are categorized by year, make, model.
[0130] The steps of preprocessing the car image dataset D are as follows:
[0131] 1.1) Extract the serial number, image name and category name of the car image.
[0132] 1.2) Delete the grayscale car images in the car image dataset D. Delete the car images whose aspect ratio is less than h in the car image dataset D. Delete the car images whose image bytes are less than Hkb in the car image dataset D.
[0133] 1.3) Unify the car image pixels in the car image dataset D into n×...
Embodiment 2
[0248] see figure 1 , a controllable car image synthesis method based on the causal flow model, which mainly includes the following steps:
[0249] 1) Obtain the original car picture data, for the data set Do preprocessing. The car picture data is Stanford car picture data, including 196 categories of 16185 pictures, each category including year, manufacturer and model.
[0250] Further, the main steps of preprocessing the car image data are:
[0251] 1.1) Extract the sequence number, picture name, category name in the data;
[0252] 1.2) Delete the grayscale image in the car image data set, the pixel ratio of length to width is less than 0.3, and the number of image bytes is less than 10kb.
[0253] 1.3) Fix the pixel size of the picture to 64×64.
[0254] 1.4) Each car image contains 15 binary attribute annotations, including car color, car size, headlights, window glass, sunroof, model, wheels, rear combination lights, doors, roof, exterior mirrors , rear windshield,...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More
PUM
Login to View More
Abstract
The invention discloses a controllable automobile image synthesis method based on a causal flow model. The method comprises the following steps: 1) acquiring original automobile image data; 2) establishing a countercurrent model p theta (x); 3) establishing a network architecture of the countercurrent model p theta (x); 4) outputting an automobile image y; 5) establishing a causal relationship network; 6) setting a supervision condition cs according to the causal relationship network, and establishing a controllable causal encoder E theta; 7) establishing a supervision module; and 8) outputting the synthesized automobile picture. On the basis of the original flow model, the causal network is added, the supervision condition can be enhanced, and then the controllable automobile image meeting the expected target is generated.
Description
technical field [0001] The invention relates to the technical field of image processing, in particular to a controllable car image synthesis method based on a causal flow model. Background technique [0002] Nowadays, the popularity of cars is getting higher and higher, the number of car ownership has increased significantly, and a large amount of car image data has been accumulated. How to analyze these data to extract useful value? For manufacturers, it is necessary to separate commodity lines and market competition; for consumers, it is necessary to clarify the direction of purchase; for society, it is necessary to facilitate management and planning. Reasonable use of data will effectively save human resources and promote the intelligent development of the automobile industry. [0003] With the continuous development of deep learning technology, the research on image processing has been very mature, but the controllable image synthesis technology is mostly used in face r...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More
Application Information
Patent Timeline
Application Date:The date an application was filed.
Publication Date:The date a patent or application was officially published.
First Publication Date:The earliest publication date of a patent with the same application number.
Issue Date:Publication date of the patent grant document.
PCT Entry Date:The Entry date of PCT National Phase.
Estimated Expiry Date:The statutory expiry date of a patent right according to the Patent Law, and it is the longest term of protection that the patent right can achieve without the termination of the patent right due to other reasons(Term extension factor has been taken into account ).
Invalid Date:Actual expiry date is based on effective date or publication date of legal transaction data of invalid patent.
Login to View More 
Login to View More