Doubt regarding val
Is the val.zip the train/val split that is used during training? Or is it the test set that the models were evaluated & reported after training on the train/val split?
Thanks for your question!
The train and val splits are only used during the training stage. Specifically, we train the model on the train split and use the val split for validation and monitoring during training.
We separate a relatively small subset as val mainly for efficiency. Using the entire training set for validation would significantly increase the validation time, while the purpose of this split is simply to monitor the training process and model convergence.
Therefore, val.zip is not the test set used for the final reported results. After training is completed, we evaluate the trained models on the other test datasets used in our experiments (https://huggingface.co/datasets/kkk2026/OVRSISBenchtest, https://huggingface.co/datasets/kkk2026/OVRSISBenchV2o3).
Hope this clarifies the dataset split!