model.train() setting
GlancerZ opened this issue · 1 comments
GlancerZ commented
Why do you use model.train in the inference part (encode batch)?
jsxlei commented
Thank you for your interest. Here, model.train means using data-specific mean and var in the batch_norm, otherwise it will use the mean and var from the training data. Anyway, the model.train or model.eval does not make big difference.