Open Yzhbuaa opened 2 weeks ago
Your advice is excellent! We have also found that the most significant factor contributing to unstable generation is - just using 4 views! We are in the process of training the model with 8 views and hope to see a significant improvement in generation stability.
Hello, thank you for your great work! Currently, the model has a batch of 4 images generated.
If we want more views, for example, a top view and a bottom view. Should we just add those 2 views to training dataset and extend both the batch size and class labels length to 6?
Thank you!