✨ BeltCrack ✨

To propel machine learning advancement in this field, we construct a pair of belt crack detection datasets sequential-image belt crack detection datasets (BeltCrack14ks and BeltCrack9kd), from real-word production factory .

[Patter Recognization] BeltCrack: the First Sequential-image Industrial Conveyor Belt Crack Detection Dataset and Its Baseline with Triple-domain Feature Learning

⭐⭐ If this work is useful to you, please give this repository a star ! ⭐⭐

📢 News

• 2025-06-23: The official Pytorch implementation for BeltCrackDet is publicly available in this repository 📦!
• 2025-06-23: Our industrial belt crack datasets are released for public 🎁!
• 2025-06-22: We have released our technical report to ArXiv, here 📤.

📊 Datasets Overview

Download

• You can get the BeltCrack14ks from Quark Disk in this link with password: fUTU
• You can get the BeltCrack9kd from Quark Disk in this link with password: E1St

Caption

• The dataset BeltCrack14ks contains 14,087 images, across 29 sequences. While BeltCrack9kd comprises 9,645 images, from 42 sequences.
• They are captured in real-world industrial environments, including conveyor the belt cracks under multiple perspectives (top-down, bottom-up), varying the lighting conditions from morning strong light to evening low illumination, extreme weather (sunny, rainy, snowy), and dynamic belt moving speeds.

🔑 Method

We propose a tri-path network architecture to implement cross-domain representation learning through Hierarchical Spatial-aware Module (HSM), Aggregative Temporal Module (ATM), and Wavelet-enhanced Frequency-aware Module (WFM). In addition, Residual Compensation Unit (RCU) dynamically mitigates inter-domain representational gaps, while optimizing cross-domain feature fusion.

🔱 Getting started

Enviroment

You can create your own conda environment for BeltCrackDet based on the following commands:

conda create -n BeltCrackDet python=3.9 
conda activate BeltCrackDet
pip install torch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 --index-url https://download.pytorch.org/whl/cu118
pip install opencv-python==4.11.0.86
pip install einops scikit-learn numpy

Train

You could modify the parameters or paths in the train_BeltCrackDet.py file and run it with the following command (for two GPUs):

CUDA_VISIBLE_DEVICES=0,1 python train_BeltCrackDet.py 

Test

Once training is complete, you could choose best model (usually not 'best_epoch_weights.pth') from "results/beltcrack" to test the performance, and use the following command (for two GPUs):

CUDA_VISIBLE_DEVICES=0,1 python vid_map_coco.py

Visulization

You could choose the mode "predict" in the "predict.py" file to get the results:

python predict.py

The visualization results of our comparative experiments are as follows:

🗓️ TODO

• [🟢 Complete] arXiv preprint release
• [🟢 Complete] Open datasets to public
• [🟢 Complete] Open source code at this repository
• [🟢 Complete] Add a description of our datastes and baseline in readme
• [🟢 Complete] Add visualization of experiment results in readme
• [🟡 In Progress] Submit original manuscripts to Pattern Recognition

🏷️License

This project is released under the Apache 2.0 license.

📧 Concat

If you have some questions about this work, you can open an issue or contact me via email (with the subject of BeltCrack): [email protected]