emvd.md

Contents

• EMVD Description
• Dataset
• Environmental requirements
• Performance
  • Training performance
  • Evaluation performance

EMVD Description

Efficient Multi-Stage Video Denoising With Recurrent Spatio-Temporal Fusion.

EMVD is an efficient video denoising method which recursively exploit the spatio temporal correlation inherently present in natural videos through multiple cascading processing stages applied in a recurrent fashion, namely temporal fusion, spatial denoising, and spatio-temporal refinement.

Paper: Accelerating the Super-Resolution Convolutional Neural Network

Reference github repository (PyTorch)

Dataset

CRVD Dataset

Environmental requirements

GPU

• Hardware (GPU)
  • Prepare hardware environment with GPU processor
• Framework
  • MindSpore
• For details, see the following resources:
  • MindSpore Tutorial
  • MindSpore Python API
• Additional python packages:
  • Install additional packages manually or using pip install -r requirements.txt command in the model directory.

Ascend 910

• Hardware (Ascend)
  • Prepare hardware environment with Ascend 910 (cann_5.1.2, euler_2.8.3, py_3.7)
• Framework
  • MindSpore Ascend 1.9.0 or later

Performance

Training Performance

Evaluation Performance

Model	Device type	Device	PSNR (dB)	Train time (secs per epoch)
EMVD-Torch	GPU	V100	42.09	29
EMVD-MS-GPU	GPU	V100	42.12	28
EMVD-MS-Ascend	Ascend	Ascend-910A	40.77	19