102flower.pytorch

This repository contains the code to train data taken from Oxford 102 Category Flower Dataset as an image classification using pytorch. I have used resnet34 pretrained model from torchvision to train the model and made some codes to train the module.

Requirment:

• Python=3.6.8
• pytorch=1.9.0
• Numpy=1.16.6
• fire=0.4.0
• torchvision=0.10.0

i have also save the package-list.txt into the repository.

Hardware Requirment:

• Computer with decend RAM and CPU
• GPU (optional)

How to Use:

Dataset:

• Download the dataset manually from here: https://www.kaggle.com/nunenuh/pytorch-challange-flower-dataset/download
• Store dataset at dataset/

Training:

• Use train.py to train the model.
• Change dataset path to the appropriate path if needed
• You can modify the Hyperparameter and Augmentation if needed
• Use this command 'python train.py --help' for help

example command:

python train.py  --lrate 0.05 --epoch 100 --lfreq 20 --bsize 128 --num_worker 64 

Test:

• Use 'test.py' to test the model that you have trained.
• modify the image_path(default: dataset/test/image_*) and weight_path(default: weiights/model_best.pth) to the specific path location to test the image
• set the download variable to 'True' if you want yo use my last trained weights or you could train it yourself (follow 'Training' steps)
• if your device have gpu set the use_gpu to 'True'
• the program will predict the class (or classes) of an image using a trained deep learning model

example command:

python test.py --image_path=dataset/valid/2/image_05094.jpg --weight_path=weights/model_best.pth --use_gpu=False

• output result:

Flower category: hard-leaved pocket orchid

Prediction

Class: hard-leaved pocket orchid , confidence: 99.997%

Class: wild pansy , confidence: 0.001%

Class: moon orchid , confidence: 0.001%

Class: japanese anemone , confidence: 0.000%

Class: lotus lotus , confidence: 0.000%

Reference:

• Oxford 102 Flower Dataset, https://www.robots.ox.ac.uk/~vgg/data/flowers/102/
• Nunenuh, Oxford 102 Flower Dataset, https://www.kaggle.com/nunenuh/pytorch-challange-flower-dataset
• ResNet, Pytorch, https://pytorch.org/hub/pytorch_vision_resnet/
• Torchvision Documentation, Pytorch, https://pytorch.org/vision/stable/index.html