[Project Addition] Rock Formation Classification using DL

[Arihant-Bhandari]

Pull Request for DL-Simplified 💡

Issue Title : #729 Rock Formation Classification using DL

• Info about the related issue (Aim of the project): Classifying rock formations into Igneous, Metamorphic and sedimentary types.
• Name: Arihant Bhandari
• GitHub ID: https://github.com/Arihant-Bhandari
• Email ID: arihant.bhandari.2004@gmail.com
• Idenitfy yourself: GSSoC 2024 Contributor

Closes: #729 Rock Formation Classification using DL

Describe the add-ons or changes you've made 📃

Wrote code for preprocessing and loading images from Dataset, turned labels into probability distributions, created models including custom CNN as well as pre-trained models, like ResNet-50, VGG16 and so on.

Type of change ☑️

What sort of change have you made:

• [x] Bug fix (non-breaking change which fixes an issue)
• [x] New feature (non-breaking change which adds functionality)
• [x] Code style update (formatting, local variables)
• [x] Breaking change (fix or feature that would cause existing functionality to not work as expected)
• [x] This change requires a documentation update

How Has This Been Tested? ⚙️

Models	Accuracy	Loss
ResNet-50	27.73%	11.6480
InceptionV3	56.09%	0.9833
CNN	54.76%	0.9459
VGG16	59.73%	0.9362
EfficientNetB7	37.51%	1.0937
DenseNet-121	61.87%	0.8813
Xception	58.76%	0.9247

Checklist: ☑️

• [x] My code follows the guidelines of this project.
• [x] I have performed a self-review of my own code.
• [x] I have commented my code, particularly wherever it was hard to understand.
• [x] I have made corresponding changes to the documentation.
• [x] My changes generate no new warnings.
• [x] I have added things that prove my fix is effective or that my feature works.
• [x] Any dependent changes have been merged and published in downstream modules.

[github-actions[bot]]

Our team will soon review your PR. Thanks @Arihant-Bhandari :)

[Arihant-Bhandari]

hi @abhisheks008 pls check this PR, if any changes to be made, will gladly do them over.

thank you for your time and patience.

[Arihant-Bhandari]

hi @abhisheks008 i have tried out various strategies, and unfortunately couldnt come up with a suitable result. i hope this issue can be sidelined for now, i will try other things on this dataset like vision transformers and stuff, but for now i hope this issue can be sidelined.

i have completed the other project , about arabic dates, and i can hand that in today. i hope you can allow me this switch.

[abhisheks008]

hi @abhisheks008 i have tried out various strategies, and unfortunately couldnt come up with a suitable result. i hope this issue can be sidelined for now, i will try other things on this dataset like vision transformers and stuff, but for now i hope this issue can be sidelined.

i have completed the other project , about arabic dates, and i can hand that in today. i hope you can allow me this switch.

Okay!

[Arihant-Bhandari]

hi @abhisheks008 , wanted to update you on this, i have implemented a MoCo model on it for self supervised learning technique, and i am straightening out rough parts, i hope this implementation of MoCo can be regarded as USP of this project.

[Arihant-Bhandari]

hi @abhisheks008 , i wanted to talk abt this project's current progress: so the model has been trained, with a updation priority queue format under Momentum Contrast technique, but there isnt an increase in the accuracy score, i am getting a score of about 52% accuracy thru it.

i wanted to ask for your opinion on this: would u allow submission of this project as it stands considering the MoCo implementation as a USP, or do we shut it down becuase there isnt that much variability for a CNN to properly capture features as far as i have seen. whatever you think i am ok with it, pls let me know your thoughts on it

[abhisheks008]

hi @abhisheks008 , i wanted to talk abt this project's current progress: so the model has been trained, with a updation priority queue format under Momentum Contrast technique, but there isnt an increase in the accuracy score, i am getting a score of about 52% accuracy thru it.

i wanted to ask for your opinion on this: would u allow submission of this project as it stands considering the MoCo implementation as a USP, or do we shut it down becuase there isnt that much variability for a CNN to properly capture features as far as i have seen. whatever you think i am ok with it, pls let me know your thoughts on it

hi @abhisheks008 , i wanted to talk abt this project's current progress: so the model has been trained, with a updation priority queue format under Momentum Contrast technique, but there isnt an increase in the accuracy score, i am getting a score of about 52% accuracy thru it.

i wanted to ask for your opinion on this: would u allow submission of this project as it stands considering the MoCo implementation as a USP, or do we shut it down becuase there isnt that much variability for a CNN to properly capture features as far as i have seen. whatever you think i am ok with it, pls let me know your thoughts on it

Yes getting an accuracy of 52% is not enough for this problem statement. Closing this issue as not reaching the threshold of accuracy score.