Training problems with the code

[ZZ2490]

Is the code provided now only for evaluation and not for training?

[Hongbin98]

Thanks for your attention to our work~ We have provided the training cmd CUDA_VISIBLE_DEVICES=0 python tools/tta_monotta.py --config runs/monoflex.yaml --ckpt model_moderate_best_soft.pth --eval --output kitti-c/gaussian1 of MonoTTA. If you have any questions, please feel free to ask.

[ZZ2490]

Running the above commands only evaluates, it does not train

[Hongbin98]

MonoTTA follows a new adaptation paradigm termed Fully Test-time Adaptation which aims to adapt a well-trained model to unlabeled test data by handling potential data distribution shifts at test time. So MonoTTA would adapt the well-trained model at the inference stage (before providing the final prediction). If you are still confused, you may check the related work (Fully Test-time Adaptation methods) in classification tasks~