Skip to content

Latest commit

 

History

History
97 lines (74 loc) · 5.07 KB

fsrcnn.md

File metadata and controls

97 lines (74 loc) · 5.07 KB
Raw

Contents

FSRCNN Description

As a successful deep model applied in image super-resolution (SR), the Super-Resolution Convolutional Neural Network ( SRCNN) [1,2] has demonstrated superior performance to the previous hand-crafted models either in speed and restoration quality. However, the high computational cost still hinders it from practical usage that demands real-time performance ( 24 fps). In this paper, we aim at accelerating the current SRCNN, and propose a compact hourglass-shape CNN structure for faster and better SR. We re-design the SRCNN structure mainly in three aspects. First, we introduce a deconvolution layer at the end of the network, then the mapping is learned directly from the original low-resolution image (without interpolation) to the high-resolution one. Second, we reformulate the mapping layer by shrinking the input feature dimension before mapping and expanding back afterwards. Third, we adopt smaller filter sizes but more mapping layers. The proposed model achieves a speed up of more than 40 times with even superior restoration quality. Further, we present the parameter settings that can achieve real-time performance on a generic CPU while still maintaining good performance. A corresponding transfer strategy is also proposed for fast training and testing across different upscaling factors.

Paper: Accelerating the Super-Resolution Convolutional Neural Network

Reference github repository (PyTorch)

Dataset

The 91-image, Set5 dataset converted to HDF5 can be downloaded from the links below.

Dataset Scale Type Link
91-image 2 Train Download
91-image 3 Train Download
91-image 4 Train Download
Set5 2 Eval Download
Set5 3 Eval Download
Set5 4 Eval Download

Environmental requirements

GPU

  • Hardware (GPU)
    • Prepare hardware environment with GPU processor
  • Framework
  • For details, see the following resources:
  • 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

Performance

Training Performance

To compare results we use scale = 4 here and everywhere in the report.

Model Device type Device PSNR (dB) Train time (secs per epoch)
FSRCNN-Torch-paper GPU ? 30.55 ?
FSRCNN-Torch-github GPU ? 30.50 ?
FSRCNN-Torch GPU P100 30.44 210
FSRCNN-MS GPU P100 30.15 320
FSRCNN-MS-Ascend Ascend Ascend-910A 30.18 360

PSNR differences between the original (paper and github) and our implementations are small. So the result are reproduced.

Evaluation Performance

Inference dataset info
Dataset name Set5
batch size 8
Scale x4
Experiments num (its) 1000
Warm up (its) 100

Timings

Model Device type Device Inference time (ms, median, per it)
FSRCNN-Torch GPU P100 6
FSRCNN-MS GPU P100 13
FSRCNN-MS-Ascend Ascend Ascend-910A 20