Questions on Model training

[aravindaideas2it]

I have multiple static scenes, and for each of these static scene I have 48 Multi-view images, I want to train the Neus2 Model using this data. I have the following questions,

1. Should I train the Neus2 for each static scene separately using their corresponding images? or Can I combine them to train the model.
2. If I can combine the training data, how should I differentiate that this set of input images belong to this particular static scene? Can you please provide some explanation on the input data convention for this scenario?

Additionally, Once I have trained the model using a static scene, How can I test the model? Is there a custom script available for testing ?

Your response to this would be greatly appreciated, Thanks in Advance.

[erin47]

Hi @aravindaideas2it , thank you for your interest in our work. Unfortunately, we currently don’t support combining all the datasets together. So, you’d better organize the dataset separately and write your own script to run them. For your next question, if you need test the model, you can try to split your training data into train&test, and render the final model using test.json to see if the renderings align with the ground truth. I hope this helps.