Question about using extrinsic in training

[1320harry]

As mentioned in issue #40, could you explain why extrinsics are being used during training? If it's supposed to be pose-free, I would assume GT extrinsics shouldn't be used during either training or testing. I'm a bit confused about this.

[zhongyingji]

During training, the model requires extrinsics for supervision on target views, the extrinsics indicate the relative positions of the target views with the first input view.
During testing, given input views, for quantitative evaluation, they optimize the extrinsics with the gt images. The extrinsics of the test views are not required, but you can use it as a starting point of the optimization.

No extrinsics or optimization are required during testing if you do not conduct the quantitative evaluation.

[1320harry]

Thanks!