feat: add MobileNetV2 Pokémon classification pipeline and notebook

[shrinidhipw05]

Issue Title : Pokémon classification using Deep learning

• Info about the related issue (Aim of the project) : To classify 150 different Pokémon species of Gen 1 based on a 7000+ labeled image dataset.
• Name: Shrinidhi Walvekar
• GitHub ID: shrinidhipw05
• Email ID: ec24bt006@iitdh.ac.in
• Identify yourself: GSSoC Participant

Closes #966

Technical Implementation Summary

I have successfully built, optimized, and finalized the end-to-end computer vision pipeline for this project. Here is a breakdown of the development steps and engineering choices made:

• Optimized Data Pipeline: Utilized image_dataset_from_directory with an 80/20 train/validation split across the 150 classes (processing 6,820 valid images). Applied .cache() and .prefetch(AUTOTUNE) performance optimization alongside on-the-fly data augmentation (RandomFlip, RandomRotation, RandomZoom) to build model robustness.
• Architecture Optimization & Pivot: Initially built a Custom CNN baseline which exhibited severe overfitting (~82% train vs. ~32% validation accuracy). After testing a heavy VGG16 architecture that bottlenecked training times on CPU, I pivoted to MobileNetV2 transfer learning (with frozen ImageNet weights and proper preprocess_input matching).
• Performance & Validation: MobileNetV2 successfully stabilized the training process, reaching ~67% validation accuracy in just 5 epochs. Training time dropped to a highly efficient ~50 seconds per epoch on CPU with excellent convergence.
• Deliverables: Integrated matplotlib learning curves directly within the notebook to visually map accuracy and loss trajectories. The final optimized model weights have been compiled and exported cleanly to `Model/pokemon_mobilenetv2.keras.

Comprehensive markdown notes have been embedded within the notebook to explain the architectural pivot, ensuring the notebook serves as a high-quality, readable asset for the repository. Ready for review!

[github-actions]

Our team will soon review your PR. Thanks @shrinidhipw05 :)

[abhisheks008]

Please follow the project folder structure mentioned in the README of the project repository. Based on that, update the files/folders and re-push your PR.
@shrinidhipw05

[shrinidhipw05]

Hi @abhisheks008, thanks for the review! I have completely updated the file structure to match the repository's template exactly:

• Moved the notebook and the .keras model weights strictly into the Model/ directory.
• Cleaned up the duplicate tracking of the notebook from the root directory.
• As per the repository guidelines for large datasets, I deliberately kept the raw 7000+ images out of the PR to prevent repository bloat. Instead, I added a README.md inside the Dataset/ folder containing the direct Kaggle link.

Let me know if anything else is needed. Ready for review!