Topic recognition - 47184530_VQVAE_Oasis_BrainMRI Project

[uqmshawn]

I have successfully completed the recognition problem as per the assessment guidelines. My solution, coded in Google Colab Pro due to local PC constraints, is based on the VQVAE using the OASIS Brain MRI dataset. I have followed the steps outlined in the checklist:

My solution includes the following main files:

• 47184530.ipynb: This included a running code and my commits are done from here.
• modules.py: Contains the model components.
• dataset.py: Data loader for data processing.
• train.py: Code for training, validating, and testing the model.
• predict.py: Demonstrates the usage of the trained model.
• README.md: Documents the entire project, including results and model summary.

I ensured to commit my progress regularly, maintaining a detailed commit log to provide evidence of originality and individual work. Datasets or model files are not committed. Only code, scripts, and content for the README.md file have been included in the commits.

[SiyuLiu0329]

This is an initial inspection, no action is required at this point

Difficulty: Hard

Readme Overall: 
Good

• Project Overview: Good
• Model: diagram for image generation is not accurate, should be PixelCNN -> VQ -> Decoder -> Generated image
• Data + Preprocessing: no description
• Usage: Good
• Training / Loss Curve(s): no graph(s)
• Result Demonstration: Recon ok, but no SSIM & missing results for image generation
• References: Good

Functionality:
- Reconstruction: Looks okay but no SSIM reporting
- Generation: Not completed

Code:

• Consistent with Results: Yes
• File Structure: Good
• Commenting: Good
• Commit frequency: could be more frequent (lump sum commits)
• Commit messages: some could be more descriptive

Other Comments:

• Important: You may have moved some of Shakes' initial files into your own folder. You should only make changes to your own folder. Please restore Shakes' initial files.
• There is code for image generation, but i am unable to locate any results for image generation.
• As per notebook training logs, image generation code was not executed before KeyboardInterruption.

[uqmshawn]

Hi Siyu, Hope you are doing well. I would like to highlight few points here regarding the image generation code. I have successfully generated my images through colab using the notebook and I am able to prove it as I have all the generative output files which I used to create the GIF. When I uploaded the notebook, I am sure that the output was available in there. Please let me know what might have happened. Could it be because the notebook was run without properly defining the path of the dataset. Rest be assured that I have used the following notebook to complete the training and generated the Output images I have used in my Readme which also includes the plots of graphs. Lastly, I have more then 60-70+ commits, unsure why there would only state "Only 1 commit". As I have spoken to multiple tutors and Shakes, regarding my issue with my computer, I did most of my code using my Mobile and different PC in UQ. I had to clone and upload my repo to PatternAnalysis-2023 Github after I completed the assignment from my repo. Please help me understand these issues as I have worked hard into completing this assignment and had to buy the computation from Google Colab, I can prove the runtime and transactions too. Thank You Meheraj Shawn

…

________________________________ From: Siyu Liu ***@***.***> Sent: Wednesday, November 8, 2023 2:29 PM To: shakes76/PatternAnalysis-2023 ***@***.***> Cc: Meheraj Shawn ***@***.***>; Author ***@***.***> Subject: Re: [shakes76/PatternAnalysis-2023] Topic recognition - 47184530_VQVAE_Oasis_BrainMRI Project (PR #149) This is an initial inspection, no action is required at this point Difficulty: Hard Readme Overall: 
Good * Project Overview: Good * Model: diagram for image generation is not accurate, should be PixelCNN -> VQ -> Decoder -> Generated image * Data + Preprocessing: no description * Usage: Good * Training / Loss Curve(s): no graph(s) * Result Demonstration: Recon ok, but no SSIM & missing results for image generation * References: Good Functionality: - Reconstruction: Looks okay but no SSIM reporting - Generation: Not completed Code: * Consistent with Results: Yes * File Structure: Good * Commenting: Good - Commit frequency: only 1 commit - Commit messages: only 1 commit Other Comments: * Important: You may have moved some of Shakes' initial files into your own folder. You should only make changes to your own folder. Please restore Shakes' initial files. * There is code for image generation, but i am unable to locate any results for image generation. * As per notebook training logs, image generation code was not executed before KeyboardInterruption. — Reply to this email directly, view it on GitHub<#149 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/BBUEWSELJVKDLWOVHD3DRR3YDMRGLAVCNFSM6AAAAAA6KUECH2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMBRGE3TINZVGI>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[SiyuLiu0329]

Hi Meheraj,

When I did the inspection, I was inside your project folder, and when I checked the commit history there, only 1 showed up (one where you moved all the files from the root dir to your project dir). Hence, I initially stated only 1 commit. However, I realised very soon after and have made corrections accordingly. You may have only received my initial comments via email (before my correction).

Regarding image generation results, if you have them, it would be best to display them in ReadMe as you did extensively for reconstruction. Notebook outputs can get overwritten and hence are not recommended for showing results.

Can you upload them and let me know?

Thanks
Siyu

[uqmshawn]

Hi Siyu,

For the image generation result the your asking to be uploaded, is it the images I have generated or the output results as in with the epochs and Loss data. I do have the text file which I used to recreate the plot (consists of Perplexities, VQ Losses and Reconstruction Losses) to be sure the plots were correct. for the output results, I will need to rerun the notebook to display the values. Would it be alright to do so or can I upload the images and the .txt file (Values are pretty jumbled in the file, i could roughly organize that). Honestly, I was not sure to also upload the output print if not I would not missed inserting them in it. most example and previous projects uploaded did not have them. If you could give me some time, I could regenerate it and upload.
Below I have uploaded the data I have used to generate plots and the output images generated.

About the commits, as mentioned earlier, I was between multiple computers and it was not easy commiting every small changes as I had to use it mainly to complete the generation process and get the program working. I did speak to few tutors about my situation. I tried my best to commit all the important bits I could.

I can confirm that I have generated this program and if you give me some time, I can regenerate it to show you the output.

Thanks for your consideration,

Shawn

Perplexities.txt
Reconstruction Loss.txt
VQ Losses.txt
O1.zip
Output_img.zip

[shakes76]

Marking

Marked as per the due date and changes after which aren't necessarily allowed to contribute to grade for fairness

Good Practice (Design/Commenting, TF/Torch Usage)

Adequate design and implementation
Good spacing and comments
Header blocks missing -1

Recognition Problem

Solves problem, but no generation -2
Driver Script present
File structure present
Shows Usage & Demo & Visualisation & Data usage
Module present
Commenting
No Data leakage
Difficulty: Hard

Commit Log

Meaningful commit messages sometimes -1
Progressive commits used

Documentation

ReadMe good
Model/technical explanation
Good Description and Comments
Markdown used and PDF submitted

Pull Request

Successful Pull Request (Working Algorithm Delivered on Time in Correct Branch)
No Feedback required
Request Description good

[shakes76]

Please revert changes to the repo README, does not affect grade only the merge.

[wangzhaomxy]

No feedback attempt and no feedback marks lost.