Open xuliuwei opened 8 months ago
In fact, we want to express that in the absence of scene consistency constraints (without fusion), our SuperFusion still has better performance than other methods. And our complete SuperFusion has the best alignment performance.
The paper says that the registration effect without fusion is better than that with fusion, how is this trained?