How to improve the speed?

[hellohaley]

Hi,

I want to speed up the training progress of the model. I only ran 32 epochs in 10 hours under GPU Tesla V100. I think it will take more time to run 500 epochs. Is this the normal training speed? (run completely according to the readme.md instance statement) what parameters can I modify to speed up the training process? Is it feasible to increase the batch size? Finally, can a command statement run the training results of multiple categories at a time instead of detecting only one category of objects at a time?

Thanks ybkscht!

[nikonnext]

Hi @hellohaley
Please tell me, did you manage to speed up the learning process?

[satpalsr]

Hi @hellohaley & @nikonnext Did you manage to speed up the training?

[julingers]

Hi @nikonnext @hellohaley @satpalsr
Did you manage to speed up the training? I trained on Tesla P100 and I changed the batch size to 16. The training progress is still very slow. Is this normal？

[satpalsr]

Hey @julingers , This may help a bit #38. But yeah the training time is still long.