mmsegmentation
mmsegmentation copied to clipboard
How to fix the seed during Multi-GPU learning
Hello!! I have some questions.
Is there a way to fix the seed number that appears in the Runtime environment information in the log every time I perform multi-GPU training? And does changing this seed number actually impact performance? I'm asking because the seed in the training logs of the models you provide keeps changing.
Hi , use randomness = dict(seed=42)
in the config to set the seed. Some models are sensitive to the seed while others are not. By default, no seed is set!