Closed Jingjinganhao closed 1 year ago
Hey @Jingjinganhao,
The model is trained to predict depth from a cost volume built with seven source images.
You could try using the same model, but provide dummy data for source views (identity for pose and the same exact intrinsics as current view). Modify the code to pass this dummy data on the forward pass.
Ideally, you would train the model without source views by ignoring features from the cost volume.
Good luck!
Hi @mohammed-amr :
Thank you for sharing . I want to run
test.py
only using one source image.Which parameters do I need to modify?
do I have to retrain the model?