Man1103/ResNet18-FT-PetImages-11.7M

License & Attribution

This repository contains weights or code derived from the ResNet18 foundational architecture developed by the ResNet Authors. This ResNet18 model is fine-tuned using popular CatsVsDogs PetImages dataset.

• —License: Distributed under the https://opensource.org/license/bsd-3-clause.
• —Original Model Base: https://huggingface.co/microsoft/resnet-18
• —Copyright Notice: Copyright 2025-2026 Hugging Face & The ResNet Authors.
• —Research paper introducing this model: https://arxiv.org/abs/1512.03385

Dataset:

You can find the fine-tuning dataset using the link: https://huggingface.co/datasets/Man1103/PetImages.

Fine-tuning Configuration Metrics:

• —Training Dataset: 50:50 class-balanced cats and dogs images with random transformations during training.
• —Training strategy: Two stage fine-tuning strategy with:
• —Stage-1: This is where only classification FC layer is trained.
• —Stage-2: This is where last two features extraction layers and classification FC layer are trained.
• —Num epochs: 30
• —Device: CUDA (Nvidia Tesla T4 GPU)
• —Training batch size: 128
• —Testing batch size: 256
• —Learning rate:
• —For stage-1: 0.001
• —For stage-2: 0.00001
• —Accuracy Score:
• —Training: 0.1 (100.0%)
• —Testing: 0.1 (100.0%)

Please refer to preprocessing and training pipelines in the model files:

• —Preprocessing pipeline: preprocessing.py
• —Training pipeline: train.py