some issues for train #3
Description
Hello, author. I trained MVGamba using the re10k dataset, which is different from the component data you provided. After training on a single card for one day to 150 epochs, it was found that the loss no longer changed after 10 epochs, at 0.58. The predicted image has almost no structure and is just a blurry area. Have you tried training scene data? How can you optimize the training to make the scene clear.