Could you please share the training code for the model? #24
Description
Dear Authors,
I am a user of your paper "Estimation of continuous valence and arousal levels from faces in naturalistic conditions" (Nature Machine Intelligence 2021). Thank you very much for open-sourcing the pre-trained models and inference-related code, which has been extremely helpful for my research.
During usage, I hope to fine-tune or retrain the model on my own dataset. However, I noticed that the current codebase mainly includes scripts for testing (test.py), single-image demonstration (demo.py), and video demonstration (demo_video.py), but no complete training code (e.g., training loops, loss function configurations, optimizer settings, etc.).
Could you kindly share the training code for the model? If you have plans to release it, could you estimate when that might be? If public release is not planned for now, would you mind briefly explaining key training parameters (such as learning rate scheduling, loss function weights, number of epochs, etc.) to guide me in implementing the training logic based on the existing modules?
Thank you very much for your time and contributions! I look forward to your reply.