Transfer Learning Vision Transformer (ViT) - Google 224 ViT Base Patch

Description

This model is a Transfer Learning Vision Transformer (ViT) based on Google's 224 ViT Base Patch architecture. It has been fine-tuned on a dataset consisting of fungal images from Russia, with a specific focus on various fungi and lichen species.

Model Information

• Model Name: Transfer Learning ViT - Google 224 ViT Base Patch
• Model Architecture: Vision Transformer (ViT)
• Base Architecture: Google's 224 ViT Base Patch
• Pre-trained on General ImageNet dataset
• Fine-tuned on: Fungal image dataset from Russia

Performance

• Accuracy: 90.31%
• F1 Score: 86.33%

Training Details

• Training Loss:
  • Initial: 1.043200
  • Final: 0.116200
• Validation Loss:
  • Initial: 0.822428
  • Final: 0.335994
• Training Epochs: 10
• Training Runtime: 18575.04 seconds
• Training Samples per Second: 33.327
• Training Steps per Second: 1.042
• Total FLOPs: 4.801 x 10^19

Recommended Use Cases

• Species classification of various fungi and lichen in Russia.
• Fungal biodiversity studies.
• Image recognition tasks related to fungi and lichen species.

Limitations

• The model's performance is optimized for fungal species and may not generalize well to other domains.
• The model may not perform well on images of fungi and lichen species from regions other than Russia.

Model Author

Siddhant Dutta