Skip to content

Latest commit

 

History

History
161 lines (142 loc) · 5.38 KB

SuperRes.md

File metadata and controls

161 lines (142 loc) · 5.38 KB

Super Resolution Family of Models

This document describes how to run AIMET quantization on the following model(s) and verify the performance of the quantized models. Example code is provided in the form of a Jupyter Notebook.

  • QuickSRNet
  • Anchor-based Plain Net (ABPN)
  • Extremely Lightweight Quantization Robust Real-Time Single-Image Super Resolution (XLSR)
  • Super-Efficient Super Resolution (SESR)

Environment setup

Clone the AIMET Model Zoo repo into your workspace:
git clone https://github.com/quic/aimet-model-zoo.git

AIMET installation and setup

Please install and setup AIMET (Torch GPU variant) before proceeding further.

NOTE

  • All AIMET releases are available here: https://github.com/quic/aimet/releases
  • This model has been tested using AIMET version 1.23.0 (i.e. set release_tag="1.23.0" in the above instructions).
  • This model is compatible with the PyTorch GPU variant of AIMET (i.e. set AIMET_VARIANT="torch_gpu" in the above instructions).

Install additional dependencies

To run the Jupyter Notebook, the following packages to be installed:

pip install jupyter==1.0.0
pip install notebook==6.4.3

When you install these packages, you should see a WARNING at the bottom, specifying to add a path to the environment variables. To do this, simply export the given path as:
export PATH=<path/from/warning/message>:$PATH

Download dataset

Download the Set14 Super Resolution Dataset from here: https://deepai.org/dataset/set14-super-resolution to any location in your workspace.

Extract the set5.zip file, then the SR_testing_datasets.zip found within.

The images of interest are located in the following path:
<root-path>/set5/SR_testing_datasets/Set14

Start Jupyter Notebook

Change to the directory containing the example code as follows: cd aimet_zoo_torch/superres

Start the notebook server as follows (please customize the command line options using the help/documentation if appropriate):
jupyter notebook --ip=* --no-browser &
The above command will generate and display a URL in the terminal. Copy and paste it into your browser.

Run Jupyter Notebook

Browse to the notebook at superres >> notebooks >> superres_quanteval.ipynb, and follow the instructions therein to run the code.

Edit Jupyter Notebook
Please replace the placeholders within the Jupyter Notebook (such as the model checkpoint and the dataset paths) to point to the appropriate paths in your workspace into where you downloaded and extracted them.

Models

Model checkpoints are available in the Releases page.

Please note the following regarding the available checkpoints:

  • All model architectures were reimplemented from scratch and trained on the DIV2k dataset (available here).
  • The float32 model is the full-precision model with the highest validation accuracy on the DIV2k dataset.
  • The int8 model is the quantized model with the highest validation accuracy obtained using AIMET's Quantization-aware Training.
  • The above quantized model along with the Encodings were exported using AIMET's export tool.

Quantization Configuration

In the evaluation notebook included, we have used the default config file, which configures the quantizer ops with the following assumptions:

  • Weight quantization: 8 bits, per tensor symmetric quantization
  • Bias parameters are not quantized
  • Activation quantization: 8 bits, asymmetric quantization
  • Model inputs are quantized
  • TF_enhanced was used as the quantization scheme

Results

NOTE: All results below used a Scaling factor (LR-to-HR upscaling) of 2x and the Set14 dataset.

Model Config[1] Channels PSNR
FP32 INT8
ABPN N/A 28 32.71 32.64
N/A 32 32.75 32.69
XLSR N/A 32 32.57 32.30
SESR M3 16 32.41 32.25
M5 16 32.57 32.50
M7 16 32.66 32.58
M11 16 32.73 32.59
XL 32 33.03 32.92
QuickSRNet Small 32 32.52 32.49
Medium 32 32.78 32.73
Large 64 33.24 33.17

[1] Config: This parameter denotes a model configuration corresponding to a certain number of residual blocks used. The Mx models have 16 feature channels, whereas the XL model has 32 feature channels.