Add ADNI ConvNeXt classifier (Problem 8)

[shiv-0831]

Overview

Implements the ADNI Alzheimer’s vs Normal classifier (Problem 8) using a ConvNeXtLite-inspired backbone, with code and documentation in recognition/adni_convnext_47280647/.

Highlights

• Subject-wise ADNI data loader with optional augmentation to remove patient leakage.
• TinyCNN baseline and ConvNeXtLite architectures exposed through modules.build_model.
• train.py and predict.py CLIs with deterministic seeding, checkpointing, and automatic curve exports.
• Comprehensive README covering dataset/methodology, results (tables + figures), usage commands, future improvements, and dependency instructions (with images stored under images/).

Key results

• Validation (subject-wise): 0.806 slice accuracy (best run).
• Held-out test: 0.653 slice accuracy / 0.689 patient accuracy (runs/rerun_lr1e-4_hd0.3_dp0.2_s42).
• Baseline subject-split rerun: 0.797 validation, 0.653 test slice accuracy / 0.667 patient accuracy (runs/rerun_subjectsplit_lr3e-4_hd0.2_s123).

Please refer to recognition/adni_convnext_47280647/README.md for plots and usage instructions.

[yexincheng]

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

Recognition Problem : total : 17.5

1. Solves problem: The solution is less appropriate for the problem, achieving an accuracy of 0.689 on the test set, which is a bit low.
   The dataset was properly split; split by subject.
   Reasonable accuracy plots but a bit weird that the validation loss kept increasing; plots for two trails provided. More hyperparameter tunning required.(3.5/5)
2. Implementation functions: Good (3/3)
3. Good design: Well-designed (1/1)
4. Commenting: No comments in train and predict scripts. (0/1)
5. Difficulty: Hard (10/10)

Note:

• All commits happened after the deadline (Friday, 31 Oct).
• Well-structured report.
• Detailed and clear instruction of the usage.

[gayanku]

Marking

Good/OK/Fair Practice (Design/Commenting, TF/Torch Usage)
	No design and implementation.	-2
	Spacing and comments.
	No Header blocks.	-1
Recognition Problem
	OK solution to problem.	-1
	Driver Script present.
	File structure present.
	Good Usage & Demo & Visualisation & Data usage.
	Module present.
	Commenting present.
	No Data leakage found.
	Difficulty : Hard. Hard. ConvNeXt
Commit Log
	Some/Adequate Meaningful commit messages.	-1
	Good Progressive commits.	-2
Documentation
	Readme :Good.
	Model/technical explanation :Good.
	Description and Comments :Good.
	Markdown used and PDF submitted.
Pull Request
	Successful Pull Request (Working Algorithm Delivered on Time in Correct Branch).
	Feedback action require: Feedback marks possible +2 if the requested changes are made. Remove chace files for merge.	-2
	Request Description is good.
TOTAL		-9

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

[wangzhaomxy]

47280647