iammartian0/vegetation_classification_model

Model description

This is a transformers based image classification model, implemented using the technique of transfer learning. The pretrained model is Vision transformer trained on Imagenet-21k.

Datasets

The dataset used is downloaded from git repo Agri-Hub/Space2Ground. I used Street-level image patches folder for this model. It is a dataset containing cropped vegetation parts of mapillary street-level images. Further details are on the linked git repo.

How to use

You can use this model directly with help of pipeline class from transformers library of hugging face


>>>from transformers import pipeline
>>>classifier = pipeline("image-classification", model="iammartian0/vegetation_classification_model")
>>>classifier(image)

uploading a target image to Hosted inference api.

Training procedure

Preprocessing

Assigining labels based on parent folder names

Image Transformations

Applied RandomResizedCrop from torchvision.transforms to all the training images.

Finetuning

Model is finetuned on the dataset for four epochs

Evaluation results

Model acheived an Top-1 accuracy of 0.929.

Further exploration to do

Trainig a multilabel model where model can find if the image is from left side or right side on top of classifying the vegetation
Fine grained classification of crop labels using Raw/Initial set of street-level images

BibTeX entry and citation info

@misc{wu2020visual,
      title={Visual Transformers: Token-based Image Representation and Processing for Computer Vision}, 
      author={Bichen Wu and Chenfeng Xu and Xiaoliang Dai and Alvin Wan and Peizhao Zhang and Zhicheng Yan and Masayoshi Tomizuka and Joseph Gonzalez and Kurt Keutzer and Peter Vajda},
      year={2020},
      eprint={2006.03677},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}


@INPROCEEDINGS{9816335,
  author={Choumos, George and Koukos, Alkiviadis and Sitokonstantinou, Vasileios and Kontoes, Charalampos},
  booktitle={2022 IEEE 14th Image, Video, and Multidimensional Signal Processing Workshop (IVMSP)},
  title={Towards Space-to-Ground Data Availability for Agriculture Monitoring},
  year={2022},
  volume={},
  number={},
  pages={1-5},
  doi={10.1109/IVMSP54334.2022.9816335}
}