Hello, please ask: the actual effect is inconsistent with the effect given by the paper. Is this a hyperparameter that needs to be adjusted？

[gangxu822]

[gangxu822]

The effect of stanford cars in the paper （all old 、new） is respectively： 53.8 | 71.9 | 45

[gangxu822]

adjust lr 0.1-》0.2，could get a good result

[xiamingtx]

Hello, I encountered the same problem. When adjusting lr to 0.2, should memax_weight be set to 1 or 2? Do all settings except lr follow the run_cars.sh script?

[gangxu822]

Hello, I encountered the same problem. When adjusting lr to 0.2, should memax_weight be set to 1 or 2? Do all settings except lr follow the run_cars.sh script? My problem is an id clustering problem, which is not suitable for using this gcd paradigm directly

[xiamingtx]

Hello, I encountered the same problem. When adjusting lr to 0.2, should memax_weight be set to 1 or 2? Do all settings except lr follow the run_cars.sh script? My problem is an id clustering problem, which is not suitable for using this gcd paradigm directly

Ok, thanks for your answer.