Ordinal-Regression-for-Beef-Grade-Classification

PyTorch Code for the paper:
"Ordinal Regression for Beef Grade Classification", ICCE 2023.
Chaehyeon Lee, Jiuk Hong, Jonghyuck Lee, Taehoon Choi and Heechul Jung.

Installation

prerequisites

• Python 3.7+
• PyTorch 1.10+
• TorchVision 0.11.2+

Details are specified in requirements.txt.

Training

We provide ordinal regression learning, hard label learning, and Gaussian-based label distribution learning.
You can change the learning method by changing --criterion that has ['CE', 'GLD', 'OR'].

The code below is an example of training using ordinal regression.

CUDA_VISIBLE_DEVICES=0 python3 main_reverse.py --model convnext_base_in22ft1k \
                                              --input_size 224 \
                                              --data_set image_folder \
                                              --data_path [path_to_train_dataset]   \
                                              --eval_data_path [path_to_test_dataset]    \
                                              --epochs 20 \
                                              --warmup_epochs 0 \
                                              --save_ckpt true \
                                              --cutmix 0 \
                                              --mixup 0 \
                                              --smoothing 0.1 \
                                              --project beef \
                                              --color_jitter 0.1 \
                                              --use_amp True \
                                              --batch_size 256 \
                                              --enable_wandb True \
                                              --drop_path 0.2 \
                                              --update_freq 2 \
                                              --criterion OR

Evaluation

Due to the limitation of GPU resources, we needed to store the predicted vectors in memory and then use them in ensemble learning.

1. Save outputs

python3 save_outputs.py \
        --data_set image_folder \
        --data_path [path_to_train_dataset] \
        --eval_data_path [path_to_test_dataset] \
        --use_amp true \
        --batch_size 8 \
        --input_size 224 \
        --eval true

2. Ensemble learning

python3 ensemble.py

Acknowledgement

This repository is built using the timm library and ConvNeXt repositories.
Based on the ConvNeXt, we implemented the ordinal regression for the beef grade classification.