Incorrect gs_out_data['images_pred']

The input image is:


I attempted to save decode_fake_mv and gs_out_data['images_pred'] in lines after 450 of VideoMV/tools/modules/unet/unet_i2vgen.py. While decode_fake_mv：

is correct, it seems that gs_out_data['images_pred']:

is not as expected, which further affects the subsequent generation process.

I attempted to visualize the Gaussian object:



but the rendered image in VideoMV/tools/modules/unet/unet_i2vgen.py does not seem to capture all the faces of the object. This leads me to suspect that the camera parameters might be incorrect.

I am using the officially recommended environment for videomv as per the documentation.

Start in VideoMV/tools/modules/unet/unet_i2vgen.py around the code after line 450, comparing decode_fake_mv with gs_out_data['images_pred'] and the Gaussian rendering shown in the report. Reproduce the issue in the officially recommended VideoMV environment, then trace the camera parameters and rendered faces until the predicted images match the expected views and subsequent generation is no longer affected.

Written by the indexing model from the issue text.