Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

使用两张3090训练最终结果是39.6258,没有达到您的40.1。请问问题可能出在哪里? #26

Open
jiajia131 opened this issue Jan 5, 2024 · 7 comments

Comments

@jiajia131
Copy link

完全使用fastinst_R50-vd-dcn_ppm-fpn_x3_640.yaml默认设置

[01/05 12:26:03 d2.evaluation.testing]: copypaste: Task: segm
[01/05 12:26:03 d2.evaluation.testing]: copypaste: AP,AP50,AP75,APs,APm,APl
[01/05 12:26:03 d2.evaluation.testing]: copypaste: 39.6315,61.6258,41.5632,17.0921,43.0538,62.6699

@jiajia131
Copy link
Author

得到40.1的结果是需要超过50个epoch吗??

@junjiehe96
Copy link
Owner

得到40.1的结果是需要超过50个epoch吗??

您好,所有模型都是训练50 epoch,batch size=16

@jiajia131
Copy link
Author

那请问您使用训练设备是?

@junjiehe96
Copy link
Owner

那请问您使用训练设备是?

4张V100和4张A100上均跑过,结果差不多

@supper-zh
Copy link

您好,同问。
我使用4张3090GPU,训练50 epoch,batch size=16,使用
fastinst_R50-vd-dcn_ppm-fpn_x3_640.yaml默认设置,遇到同样的情况:
[03/08 08:35:55] d2.evaluation.fast_eval_api INFO: Evaluate annotation type segm
[03/08 08:36:08] d2.evaluation.fast_eval_api INFO: COCOeval_opt.evaluate() finished in 12.72 seconds.
[03/08 08:36:08] d2.evaluation.fast_eval_api INFO: Accumulating evaluation results...
[03/08 08:36:09] d2.evaluation.fast_eval_api INFO: COCOeval_opt.accumulate() finished in 1.17 seconds.
[03/08 08:36:10] d2.evaluation.coco_evaluation INFO: Evaluation results for segm:

AP AP50 AP75 APs APm APl
39.6875 61.5065 41.6953 16.9503 42.9351 62.8829

@jiajia131
Copy link
Author

您好,同问。 我使用4张3090GPU,训练50 epoch,batch size=16,使用 fastinst_R50-vd-dcn_ppm-fpn_x3_640.yaml默认设置,遇到同样的情况: [03/08 08:35:55] d2.evaluation.fast_eval_api INFO:评估注释类型_segm_ [03/08 08:36:08] d2.evaluation.fast_eval_api INFO:COCOeval_opt.evaluate() 在 12.72 秒内完成。 [03/08 08:36:08] d2.evaluation.fast_eval_api INFO:正在累积评估结果... [03/08 08:36:09] d2.evaluation.fast_eval_api INFO:COCOeval_opt.accumulate() 在 1.17 秒内完成。 [03/08 08:36:10] d2.evaluation.coco_evaluation INFO:segm 的评估结果:

美联社 AP50 AP75 接入点 平均APm 磷酸盐
39.6875 61.5065 41.6953 16.9503 42.9351 62.8829

换成四张v100,可能可以

@LH-MAMBA
Copy link

LH-MAMBA commented Apr 1, 2024

作者您好,在fastinst_R50-vd-dcn_ppm-fpn_x1_576.yaml,使用两张3060,每张卡给两张图片,batch size=4,Lr=0.000025,减少了4倍,其他设置不变,AP只达到30.1312,请问还需修改什么参数提高AP?谢谢

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

4 participants