Image Classification Using Vision transformer or other model in 'timm' pretrained weight model library

Image Classification Using Vision transformer or various other model with 'timm' pretrained weight

Steps to follow:

Installation:

Anaconda Python Environment
version is working for CPU or [GPU]
Python 3.8
torchvision 0.16.1 (pip3 install torchvision==0.16.1)
torch 2.1.1 or [torch 2.1.1+cu121] (https://pytorch.org/get-started/locally/)
timm (pip3 install timm)
scikit-learn
mathplotlib

Installation (optional for image classification API):

flask (pip3 install flask)
opencv-python (pip3 install opencv-python)

Setup for demo code on EBS dataset:

1. Unzip image dataset under folder "classif_timm" folder
   
2. Run "vit_ebs_timm_gpt-4-turbo.ipynb" for the model training, model weight saving and prediction sample images
   

Running training and inferencing with other pretrained weight model which listed in the 8th cell

1. Open "timm_classif.ipynp" with Jupyter notebook
   
2. Define dataset folder in the 2nd cell
   
3. Define training hyper parameters in the 3rd cell
   
4. Define Data augmentation and normalization for training in the 4th cell
   
5. Define pretrained weight model in the 9th cell
   

Running demo API mode:

Make sure flask and opencv-python packages are installed in the python environment.

1. Open new terminal and Run "python classif_api.py"
   
2. Open new terminal and Run "python test_classif_api.py"
   

Reference:

https://timm.fast.ai/
https://github.com/huggingface/pytorch-image-models
https://youtu.be/mK0CHqLCoXA?si=UXKv2XkihnrjSp0O