Bird Fine-Grained Classification

File Directory

├── main.py
├── Net.py
├── train.py
├── train_utils.py
├── class_name_embedding.py
├── ClassnameProcessing.py
├── test.py
├── CUB_200_2011
│   ├── images
│   │   ├── 001.Black_footed_Albatross
│   │   │   ├── ***
│   │   │   └── Black_Footed_Albatross_0001_796111.jpg
│   │   ├── ***
│   │   └── 200.Common_Yellowthroat
│   │   │   ├── ***
│   │   │   └── Common_Yellowthroat_0126_190407.jpg
│   ├── attributes
│   ├── classes.txt   
│   ├── image_class_labels.txt
│   ├── train_test_split.txt
│   └── images.txt
├── data
│   ├── Bird_classes.pkl
│   ├── Bird_dictionary.pkl
│   └── Bird_glove6b_init_300d.npy
└── model
    └── model_9.pth # model name

Prerequisites

• python 3.9

pip install -r ./requirements.txt

Project Description

• This project makes algorithm to make a good performance in fine-grained classification with CUB dates on basic backbone.
• It is based on https://arxiv.org/abs/2205.10529 paper's method

Model Description

• Input : (batchsize, 3, 448, 448) image
• Output : (batchsize, 200) array (but when you test which sample image is in which class, you can ony get a class number)

Dataset

• The dataset is available at http://www.vision.caltech.edu/datasets/
• Input: birds Image
• Output: 200 size array, you can get which class of bird it is
• we only use the image of birds, not a bounding box or different information

Pre-trained Model link

• you can download each moedel at below link
• Basic ResNet101: fine tuned model with CUB dataset which is scratched in PyTorch
• Basic ResNet101 + Joint Embedding: scratched in PyTorch

Example of train & test method

1. train

• you can train the model with condition that you want

python 3 main.py -m train -k 3

2. test

• you can test the model with sample image which class of bird it is

python 3 main.py -m test -md './CUB_200_2011/images/' -sd './CUB_200_2011/images/144.Common_Tern/Common_Tern_0078_149161.jpg' 

3. test_per_class_topk_k

• you can test top_k accuracy of each class

python 3 main.py -m test_perclass -md './CUB_200_2011/images/' -sd './CUB_200_2011/images/144.Common_Tern/Common_Tern_0078_149161.jpg' -k 3

Demo Image

Reference

1. Fine-Grained Visual Classification using Self Assessment Classifier(https://arxiv.org/abs/2205.10529)