README.md

metadata

language: en
license: mit
library_name: peft
tags:
  - image-classification
  - pytorch
  - resnet
  - lora
  - birds
  - cub-200-2011
  - fine-tuning
  - computer-vision
datasets:
  - cub-200-2011
pipeline_tag: image-classification
widget:
  - src: >-
      https://images.unsplash.com/photo-1518992028580-6d57bd80f2dd?ixlib=rb-1.2.1&auto=format&fit=crop&w=600&q=80
    example_title: Example Bird 1 (e.g., Cardinal)
  - src: >-
      https://images.unsplash.com/photo-1552728089-57bdde30beb3?ixlib=rb-1.2.1&auto=format&fit=crop&w=600&q=80
    example_title: Example Bird 2 (e.g., Blue Jay)

ResNet50 + LoRA for Bird Classification (CUB-200-2011)

This repository contains LoRA (Low-Rank Adaptation) adapters fine-tuned on the CUB-200-2011 dataset for bird image classification. These adapters are designed to be applied to a standard torchvision.models.resnet50 base model.

Model Details

• Base Model: torchvision.models.resnet50 (pre-trained on ImageNet).
• Fine-tuning Method: Low-Rank Adaptation (LoRA) using the peft library.
• Dataset: Caltech-UCSD Birds-200-2011 (CUB-200-2011)
• Number of Classes: 200 bird species.
• LoRA Configuration:
  • Rank (r): 8 (as used in training, please verify/update)
  • Alpha (lora_alpha): 16 (as used in training, please verify/update)
  • Target Modules: ["fc", "conv1", "layer4.0.conv1"] (Please list the actual modules targeted during training)
  • Dropout: 0.05
  • Bias: "none"

How to Use

First, make sure you have torch, torchvision, and peft installed:

pip install torch torchvision peft Pillow