SDR-to-HDR Video Conversion with Creative Profile Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for converting Standard Dynamic Range (SDR) content to High Dynamic Range (HDR) often fail to maintain the unique creative styles and intent of the original content, leading to unsatisfactory results, especially when using machine learning algorithms that rely on generic training sets.
Innovation Solution
A method using artificial intelligence and machine-learning algorithms that incorporate creative profiles, such as those of directors, cinematographers, or colorists, to convert SDR content to HDR while preserving the intended creative style by training on specific stylistic elements and preferences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If generic machine learning algorithms are used for SDR to HDR conversion, then conversion speed and automation are improved, but creative style and intent are lost
Solution Approach 1:
The patent segments the training process into multiple specialized neural networks, each trained on content from a specific creative profile (director, cinematographer, colorist). This segmentation allows each network to preserve the unique stylistic characteristics of that creative professional while maintaining automated conversion capabilities.
Solution Approach 2:
The system changes the training parameters of the neural network by using creatively-graded SDR-to-HDR training sets specific to each creative profile. This parameter change enables the algorithm to adapt to different creative styles while maintaining automation, resolving the contradiction between speed and style preservation.
2Loss of information
If manually graded HDR is used, then creative intent is preserved, but processing time and cost increase significantly
Solution Approach 1:
The patent creates copies of creative styles by training neural networks on existing creatively-graded content. Once trained, these networks can automatically apply the creative intent to new SDR content, preserving the artistic vision while eliminating the need for time-consuming manual grading of each new piece of content.
Solution Approach 2:
The system performs preliminary action by pre-training the neural networks on creatively-graded training sets before actual conversion is needed. This advance preparation allows the networks to automatically preserve creative intent during conversion without requiring manual intervention at the time of conversion, thus reducing processing time.
3Measurement precision
If deep neural networks are trained with large datasets, then conversion accuracy improves, but unique creative styles are averaged out
Solution Approach 1:
Instead of using one large generic dataset, the patent segments the training data into multiple smaller datasets, each associated with a specific creative profile. This segmentation allows the system to maintain high conversion accuracy within each style category while preserving the unique characteristics of each creative professional's work.
Solution Approach 2:
The system applies local quality by creating specialized neural networks for different creative profiles rather than a single generic network. Each network is optimized for its specific creative style, allowing high accuracy within that style's domain while maintaining the unique local characteristics of each creative approach.
Data Source
Figure 1
Figure 2
Figure 3A~3B
AI summary
A method for converting a source video content constrained to a first color space to a video content constrained to a second color space using an artificial intelligence machine-learning algorithm based on a creative profile.