Crop Image Recognition

This repository is code of AI CUP 2022 Fall Crop Image Recognition Competition. We achieved a public ranking of 9th and a private ranking of 8th, corresponding to scores of 0.9328596 and 0.9344163, respectively.

Getting the code

You can download all the files in this repository by cloning this repository:

git clone https://github.com/Jia-Wei-Liao/Crop_Classification.git

Folder Structure

   ├── checkpoint (Please download from Google Drive)
   │   │   └── ...
   │   │     
   ├── dataset (Please download from AI CUP Competition)
   │   │   ├── mask
   │   │   │   └── ...   
   │   │   ├── train
   │   │   │   └── ... 
   │   │   ├── public
   │   │   │   └── ... 
   │   │   ├── private
   │   │   │   └── ... 
   │   │   ├── fold_0.json
   │   │   ├── ...
   │   │   ├── fold_5.json
   │   │   ├── public.json
   │   │   └── public_and_private.json
   │   │    
   ├── script
   │   ├── debug.sh
   │   ├── infer.sh
   │   ├── moniter.sh
   │   └── train.sh
   │
   ├── src
   │   ├── builder.py
   │   ├── constant.py
   │   ├── dataset.py
   │   ├── logger.py
   │   ├── losses.py
   │   ├── metric.py
   │   ├── models.py
   │   ├── scheduler.py
   │   ├── trainer.py
   │   ├── transforms.py
   │   ├── tta.py
   │   └── utils.py
   │
   ├── submission
   │   └── ...
   │
   ├── generate_json_file.py
   ├── generate_merge_csv.py
   ├── inference.py
   ├── requirement.txt
   └── train.py

Setting the environment

To set the environment, you can run this command:

conda create --name crop_cls python=3.7.4
source activate crop_cls
pip install -r requirements.txt

Download the dataset

You can download the dataset from the aidea's website:
https://aidea-web.tw/topic/5f632f38-7213-4d4d-bea3-117ff13c1afb

Download the checkpoint

You can download the checkpoint from our Google Drive:
https://drive.google.com/drive/folders/1cZTBzgOuuf3ms6V__71nWlOXtvmT0TxZ?usp=sharing

Training

To train the model, you can run this command:

bash script/train.sh

or

python train.py -ep <epoch> \
                -bs <batch size> \
                --model <model> \
                --fold <k fold validation data> \
                --image_size <crop and resize to img_size> \
                --loss <loss function> \
                --optim <optimizer> \
                --lr <learning rate> \
                --weight_decay <parameter weight decay> \
                --scheduler <learning rate schedule> \
                --auto_aug <use auto-augmentation> \
                --num_workers <number worker> \
                --device <gpu id>

Inference

To get the inference results, you can run this command:

bash script/infer.sh

or

python inference.py --checkpoint <MONTH-DAY-HOUR-MIN-SEC> \
                    --topk <number of model you want to ensemble>

Reproducing submission

To reproduce our submission, please do the following steps:
STEP 1: Getting the code
STEP 2: Setting the environment
STEP 3: Download the dataset
STEP 4: Download the checkpoint
STEP 5: Run the command to get the submissions:

bash reproduce.sh

STEP 6: Inference for the yenjia's repository and move the csv result to submission directory
STEP 7: Run the following command to merge all the csv files, then you can get the result.csv file on the submission folder (according to the file in submission directory)

python generate_merge_csv.py

Acknowledgement

We thank the another teammate yenjia's repository:
https://github.com/yenjia/AIdea_crops

Operating System and Device

We develop the code on Ubuntu 22.04 operating system and use python 3.9.15 version. All trainings are performed on a server with a single NVIDIA 3090 GPU.

Citation

@misc{
    title  = {crop_classification},
    author = {Jia-Wei Liao, Yen-Jia Chen, Yi-Cheng Hung, Jing-En Hung, Shang-Yen Lee},
    url    = {https://github.com/Jia-Wei-Liao/Crop_Classification},
    year   = {2022}
}