README.md

metadata

tags:
  - biology
  - pytorch
metrics:
  - accuracy
pipeline_tag: image-classification

Butterfly image classification model that use pre-trained cnn model resnet18 and fine-tuned the last fully connected layer to classify 75 categories of butterfly species. The model used the best checkpoint with 90% test accuracy. The model constructed on Pytorch environment.

training and testing result:

Epoch: 28 Train Loss: 0.17 Train Accuracy: 0.96 Test Accuracy: 0.90

to use this model you have to:

1. download the model
2. load pretrained model resnet18
3. model = models.resnet18(pretrained=True)
4. load checkpoint from your local
5. checkpoint = torch.load('butterfly_resnet_checkpoint.model')
6. model_for_predict.load_state_dict(checkpoint)
7. predict the images
8. model_for_predict.eval())