https://github.com/sameetasadullah/cnn-benchmark-suite

A modular deep learning evaluation framework for benchmarking multiple CNN architectures across varied optimization strategies and training configurations. Built for scalable experimentation and transferability to real-world image classification tasks.
https://github.com/sameetasadullah/cnn-benchmark-suite

alexnet cifar10 cnn computer-vision deep-learning googlenet hyperparameter-tuning image-classification mobilenet model-benchmarking model-evaluation pytorch resnet torchvision

Last synced: 28 days ago
JSON representation

A modular deep learning evaluation framework for benchmarking multiple CNN architectures across varied optimization strategies and training configurations. Built for scalable experimentation and transferability to real-world image classification tasks.

• Host: GitHub
• URL: https://github.com/sameetasadullah/cnn-benchmark-suite
• Owner: SameetAsadullah
• Created: 2024-11-10T06:36:42.000Z (11 months ago)
• Default Branch: main
• Last Pushed: 2025-06-19T11:21:46.000Z (4 months ago)
• Last Synced: 2025-06-19T12:29:51.164Z (4 months ago)
• Topics: alexnet, cifar10, cnn, computer-vision, deep-learning, googlenet, hyperparameter-tuning, image-classification, mobilenet, model-benchmarking, model-evaluation, pytorch, resnet, torchvision
• Language: Jupyter Notebook
• Homepage:
• Size: 806 KB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# 🔍 Vision Model Evaluation Framework (PyTorch)

A scalable framework for benchmarking deep convolutional architectures on classification tasks. This project evaluates the impact of optimizers, learning rates, and batch sizes across multiple CNN backbones — providing a reproducible experimental setup aligned with research and production best practices.

---

## 🚀 Key Highlights

- Plug-and-play support for multiple CNN architectures

- Scalable benchmarking with grid search over:

  - Optimizers: `SGD`, `Adam`

  - Learning Rates: `0.01`, `0.1`

  - Batch Sizes: `32`, `64`

- Data augmentation, normalization, and stratified validation split

- Detailed metrics logging + real-time visualization support

- Model weights saved automatically for top-performing configs

---

## 🧠 Architectures Supported

- **Custom Lightweight CNN** (baseline)

- **ResNet-18**

- **MobileNetV2**

- **GoogleNet**

- **AlexNet** *(included for completeness; not recommended for production)*

> 🛡️ Modular design allows easy plug-in of ViT, EfficientNet, ConvNext, etc.

---

## 📊 Experiments & Logging

- Validation and test performance tracked across all combinations

- Key metrics:

  - Train & Val Accuracy/Loss (per epoch)

  - Best Validation Accuracy (per config)

  - Final Test Accuracy (per model)

- Automatic selection of best model per architecture

- Visual analytics:

  - Accuracy trends

  - Impact of learning rate

  - Batch size comparison

  - Model vs Optimizer performance

---

## 🔁 Use Cases

- Model architecture benchmarking

- Optimizer sensitivity studies

- Lightweight deployment model search

- Academic reproducibility experiments