rr8shah/TSARA

Official PyTorch implementations of our ICASSP 2021 paper "REAL VERSUS FAKE 4K - AUTHENTIC RESOLUTION ASSESSMENT"

TSARA

This is the official PyTorch implementations of our ICASSP 2021 paper "Real Versus Fake 4K - Authentic Resolution Assessment"

• 1. Brief Introduction
  • 1.1 Backgrounds
  • 1.2 Contributions
  • 1.3 Results
  • 1.4 Citation
• 2. Dataset
• 3. Prerequest
  • 3.1 Environment
  • 3.2 Packages
  • 3.3 Pretrained Models
• 4. Running the code
  • 4.1 Python Demo for testing a single image
• 5. Codes for comparing models
• 6. License

1. Brief Introduction

1.1 Backgrounds

• In recent times, streaming 4K/UHD or even higher resolution image/video content has been increasing steadily because of the potential to deliver crisp and detail rich quality-of-experience (QoE) to end-users.
• In practice, however, the pipeline of video acquisition, production, postproduction, and delivery often involves stages where video frames are scaled down to lower resolutions, and then upscaled back to 4K/UHD resolution at later stages. As a result, the authentic 4K resolution has been lost in the process while end-users are often poorly informed of such quality degradations. Therefore, we need True/Fake 4K decision in practical applications to ensure detail rich quality-of-experience (QoE).

1.2 Contributions

• We introduced one of the first, the largest, and the only public dataset for real vs fake 4K image detection, which contains 10,824 True and Fake4K images.
• We proposed highly efficient DNN based TSARA (Two Stage Authentic Resolution Assessment) algorithm that can classify the image based on its native resolution in real time.
  • First stage of algorithm - A CNN model is used to predict the class labels of the local patches.
  • Second stage of algorithm - Patch level label predictions are aggregated and a logistic regression on detection frequency is used to make an overall assessment of the whole image or video frame.

1.3 Results

• Evaluation results on the proposed dataset

• Confidence map using local patch level prediction

1.4 Citation

2. Dataset

The dataset has been published here https://zenodo.org/record/4526657

3. Prerequest

3.1 Environment

The code has been tested on Ubuntu 18.04 with Python 3.8 and cuda 10.2

3.2 Packages

pytorch=1.3, torchvision=0.4, scikit-learn, pandas, pillow (or pillow-simd)

3.3 Pretrained Models

• Pretrained model could be found in folder pretrained_model/

4. Running the code

• This section only shows basic usages, please refer to the code for more options.

4.1 Python Demo for testing a single image

• Please make sure to open python file and mention correct image name and local image path in variable.

python demo.py

• The output should be 'True4K' for given input image.

5. Codes for comparing models

For other model compared in the paper, you can find the code in

1. FQPath: https://github.com/mahdihosseini/FQPath
2. HVS-MaxPol: https://github.com/mahdihosseini/HVS-MaxPol
3. Synthetic-MaxPol: https://github.com/mahdihosseini/Synthetic-MaxPol
4. LPC-SI: https://ece.uwaterloo.ca/~z70wang/research/lpcsi/
5. GPC: http://helios.mi.parisdescartes.fr/~moisan/sharpness/
6. MLV: https://www.mathworks.com/matlabcentral/fileexchange/49991-maximum-local-variation-mlv-code-for-sharpness-assessment-of-images
7. SPARISH: https://www.mathworks.com/matlabcentral/fileexchange/55106-sparish