Detection Accuracy Drops to 0 When Reproducing Mean Teacher & DRU – Final Code Version or My Configuration Issue? #3
Description
Hi there,
I’m currently working on reproducing the Mean Teacher and DRU modules from your codebase. However, I’ve run into a critical issue: during training/inference, the detection accuracy keeps decreasing continuously until it eventually drops to 0.
I’m writing to seek clarification on two points:
Have you uploaded the final, fully functional version of the code for Mean Teacher and DRU (e.g., are there any uncommitted updates or adjustments that might affect accuracy)?
If the code is complete, could this accuracy drop issue be caused by a configuration mistake on my end (e.g., hyperparameters, data preprocessing, or environment setup)?
Any guidance, debugging tips, or clarification on the code version would be greatly appreciated!
Thank you in advance.