I'm currently working on reproducing your "Gradient-guided-Modality-Decoupling" project and truly appreciate your open-source contribution—your work has been very insightful to my research.
However, I've encountered a gap between my reproduced results and the performance reported in your paper. I've tried adjusting several hyperparameters based on the project documentation, but still haven't been able to achieve the expected outcomes.
Given that hyperparameter settings and pre-trained weights are critical for consistent reproducibility, I'm writing to politely ask if you could share the detailed hyperparameters (e.g., learning rate, batch size, regularization parameters) and pre-trained model weights used in your original experiments. This would be incredibly helpful for me to accurately replicate your results, deepen my understanding of your method, and support my follow-up research built on your work.
If sharing the weights directly in this issue is inconvenient, feel free to contact me via email at 3077066784@qq.com.
Thank you so much for your time and effort in maintaining this project. I really appreciate any guidance or resources you can provide!
Best regards
I'm currently working on reproducing your "Gradient-guided-Modality-Decoupling" project and truly appreciate your open-source contribution—your work has been very insightful to my research.
However, I've encountered a gap between my reproduced results and the performance reported in your paper. I've tried adjusting several hyperparameters based on the project documentation, but still haven't been able to achieve the expected outcomes.
Given that hyperparameter settings and pre-trained weights are critical for consistent reproducibility, I'm writing to politely ask if you could share the detailed hyperparameters (e.g., learning rate, batch size, regularization parameters) and pre-trained model weights used in your original experiments. This would be incredibly helpful for me to accurately replicate your results, deepen my understanding of your method, and support my follow-up research built on your work.
If sharing the weights directly in this issue is inconvenient, feel free to contact me via email at 3077066784@qq.com.
Thank you so much for your time and effort in maintaining this project. I really appreciate any guidance or resources you can provide!
Best regards