Hi, I would like to ask for clarification about the expected training strategy for Task 2 Novel View Synthesis. For the final Docker submission, should participants use one joint model that handles all provided test sequences, or is it allowed for the Docker pipeline to train or optimize a separate model for each sequence during evaluation? Could you please confirm whether per-sequence optimization is allowed, as long as only the allowed source-view data from endoscope2 and the provided camera information are used, the held-out endoscope1 target images are not accessed during training or inference, and the final output follows the required format? Also, are there any runtime or storage constraints we should be aware of if the Docker performs sequence-specific optimization during evaluation? Thank you. Best, Mengting