A Novel Video Compression Approach Based on Two-Stage Learning
Author Information
Author(s): Shao Dan, Wang Ning, Chen Pu, Liu Yu, Lin Lin
Primary Institution: School of Computer Science and Technology, Changchun University, Changchun, China
Hypothesis
Can a bidirectional coding video compression model based on two-stage learning improve video compression efficiency?
Conclusion
The DeepBiVC model outperformed existing video compression methods, achieving a PSNR of 37.16 and an MS-SSIM of 0.98 on the VUG dataset.
Supporting Evidence
- DeepBiVC achieved a PSNR of 37.16 and an MS-SSIM of 0.98 on the VUG dataset at bpp = 0.3.
- The model outperformed state-of-the-art video compression methods in terms of PSNR and MS-SSIM metrics.
- Experimental results indicated significant improvements in compression efficiency while maintaining high video quality.
Takeaway
This study created a new way to compress videos that makes them smaller without losing quality, using a special two-step process.
Methodology
The study used a two-stage learning strategy for video compression, involving an image compression module and a video compression module based on optical flow estimation.
Limitations
The model's inference speed is slow, the training dataset is limited, and the model is large, making it challenging for deployment on embedded devices.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website