Why rescale_grad is set to 1.0 in train_end2end.py?

msracver / Flow-Guided-Feature-Aggregation

Flow-Guided Feature Aggregation for Video Object Detection

MIT License

723 stars 190 forks source link

Why rescale_grad is set to 1.0 in train_end2end.py? #42

Open kindloaf opened 5 years ago

kindloaf commented 5 years ago

It seems to me the rescale_grad should be set to 1/batch_size. In case of 4 GPUs training rescale_grad should be 0.25. Why is the value set to "1.0" in train_end2end.py?