Issues with Model testing.

[NishanthSV]

Hi,

I have trained the NeuS2 model incrementally with three different image datasets using the following process:

1. Step 1: Trained image_data1 using the base NeuS2 model for 3,000 iterations, saving the output files, including the checkpoint file (3000.msgpack).
2. Step 2: Trained image_data2 by loading 3000.msgpack using the --load_snapshot argument in the CLI for another 3,000 iterations, saving the output files, including the checkpoint file (6000.msgpack).
3. Step 3: Trained image_data3 by loading 6000.msgpack using the --load_snapshot argument in the CLI for another 3,000 iterations, saving the output files, including the final checkpoint file (9000.msgpack).

Now, I want to test the model using the 9000.msgpack checkpoint by providing a single image to reconstruct the 3D model.

My Questions:

1. What arguments should I pass while running run.py for testing?
2. Does the testing process require any modifications to the script?

If there are specific requirements for testing after incremental fine-tuning, please let me know. Any guidance would be greatly appreciated.

Thanks in advance!

[19reborn]

Hi, thank you for trying NeuS2. I'm a bit confused by 'providing a single image for testing'. Do you mean rendering the novel view on given camera view or finetuning trained model on that view with GT image? But both procedure can be achieved by using that single image to creat a new dataset in NeuS2 format.