YUV420-ResearchInternship

YUV420 Compatibility with CompressAI to the Hyperprior Model

To use this:

1. Create a New Virtual Enviornment and Have a fresh install of the Original CompressAI library: https://github.com/InterDigitalInc/CompressAI

2. Within Newly Created CompressAI library location. Replace the "Utils.py" and "init.py" Files in the ./compressai/datasets directory with appropriate files in Git "changes" folder

3. Implement similar change in the CompressAI library location: ./compressai/models Directory for the files "init.py" and "priors.py" with appropriate files in Git "changes" folder

4. Use Python file QualityTest.py to implement models for single image compression -> reconstruction

OR

5. Use Python file TestWithOwnNetwork.py to train own model based on YUV Images.

Note: PSNRCALC.py is run seperately

For Checkpoints:

https://drive.google.com/drive/folders/1VvpAMCk60HIhphbFVlDgWp0IlxH_JL2-?usp=sharing

---

How to run QualityTest.py

---

Run quality test with the following input parameters:

#dataset directory. Be sure to create a folder with the name "test". In this folder put the SINGLE image you wish to compress

-d C:\Users\path\to\Compressable_Image_folder

#type of Device to be used. cpu or cuda

--cuda

#Training type/ Compression type. 1 = Separate Paths, 2 = PixelShuffle

--training_type 1

#Lambda that your checkpoint/model is associated with

--lambda

#Full path and file name of the reconstructed Image

--FinalFileName C:\Users\Path\to\Directory\ReconstructedImageName.yuv

#Location of the checkpoint you wish to use.

--checkpoint C:\Users\Path\To\Desired\Checkpoint\Directory\checkpoint_best_loss.pth.tar

---

How to run TestWithOwnNetwork.py

---

#Same procedure as with the original compressAI documentation except for the following inclusion:

--training_type 1 # Training type/ Compression type. 1 = Separate Paths, 2 = PixelShuffle