Abstract:Video color style transfer aims to transform the color style of an original video by using a reference style image. Most existing methods employ neural networks, which come with challenges like opaque transfer processes and limited user control over the outcomes. Typically, users cannot fine-tune the resulting images or videos. To tackle this issue, we introduce a method that predicts specific parameters for color style transfer using two images. Initially, we train a neural network to learn the corresponding color adjustment parameters. When applying style transfer to a video, we fine-tune the network with key frames from the video and the chosen style image, generating precise transformation parameters. These are then applied to convert the color style of both images and videos. Our experimental results demonstrate that our algorithm surpasses current methods in color style transfer quality. Moreover, each parameter in our method has a specific, interpretable meaning, enabling users to understand the color style transfer process and allowing them to perform manual fine-tuning if desired.
Computer Vision and Pattern Recognition,Neural and Evolutionary Computing,Image and Video Processing
What problem does this paper attempt to address?
This paper attempts to solve several key problems in video color style transfer:
1. **Visual Continuity Problem**: Existing neural - network - based methods, when dealing with video color style transfer, are prone to causing visual incoherence, that is, the style transformation between video frames may flicker or be inconsistent.
2. **User Control Problem**: In the traditional neural - network methods for color style transfer, the process is usually opaque, and users cannot fine - tune the generated results, which limits the users' ability to control the final effect.
3. **Efficiency Problem**: Although some methods such as NLUT improve the efficiency of color style transfer by predicting 3D lookup tables, these methods still hide the specific color adjustment parameters, making it difficult for users to understand the whole process and unable to perform manual fine - tuning.
To solve these problems, the paper proposes a new neural - network - based color style transfer method, which can automatically predict the specific parameters of color style transfer, thereby achieving the following goals:
- **Improve Transparency**: By predicting specific color adjustment parameters, the process of color style transfer becomes more transparent, and users can understand the meaning of each parameter and make manual adjustments as needed.
- **Enhance User Control**: Allows users to fine - tune after color style transfer, improving the users' ability to control the final effect.
- **Enhance the Consistency of Video Style Transfer**: Through Test - On - Time Training technology, fine - tuning is carried out for specific video content and style images, improving the effect and consistency of video style transfer.
- **Improve Efficiency**: By converting the predicted color adjustment parameters into 3D LUT, ultra - high - speed color style transfer is achieved while maintaining high - quality conversion results.
In short, this paper aims to solve the visual continuity, user control and efficiency problems existing in the existing video color style transfer methods through a new neural - network method, and provide a more flexible, efficient and user - friendly solution.