search

DanielCoelho112 / synfeal

Synfeal: A Data-Driven Simulator for Camera Localization
9 stars 1 forks source link

Time limitations to train all the models #98

Closed andrefdre closed 1 year ago

andrefdre commented 1 year ago

Due to time limitations, it's not possible to train 5 models for each case and network. Currently, I only have available 1 GPU in deeplar. What should I do, @miguelriemoliveira?

DanielCoelho112 commented 1 year ago

As I previously suggested, I think that one run for each case and network is sufficient. The tables show that there are no significant differences between runs.

miguelriemoliveira commented 1 year ago

Right. Let's do 1 run per approach. Later if we have more gpus we can go back to this ...

miguelriemoliveira commented 1 year ago

@DanielCoelho112 , can't we give some priority to @andrefdre since he's close to finishing the thesis?

DanielCoelho112 commented 1 year ago

@DanielCoelho112 , can't we give some priority to @andrefdre since he's close to finishing the thesis?

Yes, we can. I'll liberate one GPU by the end of the day. Still, I think one run is enough. All papers in localization (and in Deep Learning in general) are only trained once. Running 5 runs, would take around 50 experiments, which would take too much time.

miguelriemoliveira commented 1 year ago

Right. One is enough. Just i case he needs it for some other test...

On Thu, May 4, 2023, 6:02 PM Daniel Coelho @.***> wrote:

@DanielCoelho112 https://github.com/DanielCoelho112 , can't we give some priority to @andrefdre https://github.com/andrefdre since he's close to finishing the thesis?

Yes, we can. I'll liberate one GPU by the end of the day. Still, I think one run is enough. All papers in localization (and in Deep Learning in general) are only trained once. Running 5 runs, would take around 50 experiments, which would take too much time.

— Reply to this email directly, view it on GitHub https://github.com/DanielCoelho112/synfeal/issues/98#issuecomment-1535097319, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACWTHVQR6DDPZYCRMHE35ZTXEPOMDANCNFSM6AAAAAAXVUTAHM . You are receiving this because you were mentioned.Message ID: @.***>

andrefdre commented 1 year ago

Thanks, I will then train only one model, with 2 GPU I will end the trains faster.