PBD: Plastic Bottle Dataset for Defect Detection

Zhenyuan Lin, Danhua Liu*, Lai Wei, Yubo Dong

PBD

• Plastic Bottle Body : 500 images with four common defect types (stain, concave, score and squeeze).
• Plastic Bottle Shoulder : 3,463 images with two common defect types (stain and no_burr) and one bottle neck positioning marker (pipe).

Figure 1: Some sample images from the Plastic Bottle Body dataset along with their corresponding annotations

Figure 2: Distribution of defect annotation features for the Plastic Bottle Body dataset

Figure 3: Some sample images images from the Plastic Bottle Shoulder dataset along with their corresponding annotations

Figure 4: Distribution of defect annotation features for the Plastic Bottle Shoulder dataset

Usage

This dataset is allowed for academic purposes only.

Code Configuration

This repository is modified from MMDetection. The original MMDetection repository can be found at https://github.com/open-mmlab/mmdetection.

1. Installation

• Anaconda

conda create -n PBD python=3.8

• Pytorch

# conda
conda install pytorch==2.0.0 torchvision==0.15.0 torchaudio==2.0.0 pytorch-cuda=11.7 -c pytorch -c nvidia
# or pip
pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1

• Other Extra Dependencies

# for mmcv, you can see in https://mmcv.readthedocs.io/zh-cn/latest get_started/installation.html 
pip install mmcv==2.0.0 -f https://download.openmmlab.com/mmcv/dist/cu117/torch2.0/index.html 

# for mmdetection
pip install -r requirements.txt
pip install -v -e .  # or "python setup.py develop"

2. Prepare Dataset

The PBD Dataset can be downloaded from the Baidu Cloud | Google Drive | Quark Cloud

After downloading and extracting the dataset, place it in the PBD/dataset folder.

3. How to RUN?

To start training with PBD:

# In PBD folder
python tools/train.py {config_file}
# Example:
# python configs/atss/atss_r50_fpn_1x_coco.py

4. Acknowledgment

This work is forked from MMdetection Repository https://github.com/open-mmlab/mmdetection.

Citation

@inproceedings{PBD,
  author    = {Zhenyuan Lin,Danhua Liu,Lai Wei, Yubo Dong}
  title     = {PBD: Plastic Bottle Dataset for Defect Detection}
  booktitle = {ICASSP}
  year      = {2025}
}