Unable to Reproduce Reported Results on DUO Dataset

[XJWang628]

Hi authors,

Thank you for your great work and for releasing the code of UnitModule!

I'm trying to reproduce the results reported in your paper "UnitModule: A Lightweight Joint Image Enhancement Module for Underwater Object Detection" on the DUO dataset using the provided configuration (configs/fcos/unitmodule_fcos_r50_1x_duo.py) and training script.

However, after training with the default settings (2 GPUs, 12 epochs), the obtained mAP is significantly lower than the value reported in the paper. I strictly followed the installation and data preparation instructions, and my environment matches the specified versions:

• Python 3.9
• PyTorch 1.11.0+cu113
• MMCV 2.0.0, MMEngine 0.7.4, MMDetection 3.0.0

Could you please help clarify:

1. Are there any additional training tricks or hyperparameters not included in the public config?
2. Did you use any specific random seed or data augmentation settings that might affect performance?

Any guidance would be greatly appreciated! Thank you again for your contribution.

Best regards,

[XJWang628]

The corresponding configuration files and training results are as follows.

unitmodule_fcos_r50_1x_duo.py

20260120_213329 [5].log