Multi card parallel

[Re-dot-art]

May I ask if there is a parallel issue with multiple cards here? Is there any solution? I used four v100 graphics cards to run.

[Adamdad]

In the paper, we train all models with Distributed Data Parallel with 8xV100 GPU, so parallel training should be OK. Can you check your own environment and pytorch distributed config?

[Code-of-Liujie]

[Adamdad]

Please indicate which config you are using. For all configs in this repo, we have model.train_cfg. For example

ConsistentTeacher/configs/consistent-teacher/consistent_teacher_r50_fpn_coco_180k_10p.py

Lines 52 to 58 in b49a9ce

	train_cfg=dict(
	assigner=dict(type='DynamicSoftLabelAssigner', topk=13, iou_factor=2.0),
	alpha=1,
	beta=6,
	allowed_border=-1,
	pos_weight=-1,
	debug=False),

[Code-of-Liujie]

原始邮件 发件人："Xingyi Yang"< ***@***.*** &gt;; 发件时间：2024/1/17 18:56 收件人："Adamdad/ConsistentTeacher"< ***@***.*** &gt;; 抄送人："Code-of-Liujie"< ***@***.*** &gt;;"Comment"< ***@***.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Please indicate which config you are using. For all configs in this repo, we have model.train_cfg. For example https://github.com/Adamdad/ConsistentTeacher/blob/b49a9ce1c0ae78d13528be0a9a4eff65a627d581/configs/consistent-teacher/consistent_teacher_r50_fpn_coco_180k_10p.py#L52-L58 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***&gt;

[Code-of-Liujie]

Do you want to try it remotely? 原始邮件 发件人："Xingyi Yang"< ***@***.*** &gt;; 发件时间：2024/1/17 18:56 收件人："Adamdad/ConsistentTeacher"< ***@***.*** &gt;; 抄送人："Code-of-Liujie"< ***@***.*** &gt;;"Comment"< ***@***.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Please indicate which config you are using. For all configs in this repo, we have model.train_cfg. For example https://github.com/Adamdad/ConsistentTeacher/blob/b49a9ce1c0ae78d13528be0a9a4eff65a627d581/configs/consistent-teacher/consistent_teacher_r50_fpn_coco_180k_10p.py#L52-L58 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***&gt;

[Adamdad]

Do you want to try it remotely? 原始邮件 发件人："Xingyi Yang"< @.*** >; 发件时间：2024/1/17 18:56 收件人："Adamdad/ConsistentTeacher"< @.*** >; 抄送人："Code-of-Liujie"< @.*** >;"Comment"< @.*** >; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Please indicate which config you are using. For all configs in this repo, we have model.train_cfg. For example

ConsistentTeacher/configs/consistent-teacher/consistent_teacher_r50_fpn_coco_180k_10p.py

Lines 52 to 58 in b49a9ce

	train_cfg=dict(
	assigner=dict(type='DynamicSoftLabelAssigner', topk=13, iou_factor=2.0),
	alpha=1,
	beta=6,
	allowed_border=-1,
	pos_weight=-1,
	debug=False),

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

Sorry, but what do you mean by "try it remotely"?

[Code-of-Liujie]

_base_ = './consistent_teacher_r50_fpn_coco_180k_10p.py' fold = 1 percent = 10 data = dict( train=dict( sup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}.json", ), unsup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}-unlabeled.json", ), ), ) log_config = dict( _delete_=True, interval=50, hooks=[ dict(type="TextLoggerHook", by_epoch=False), dict( type="WandbLoggerHook", init_kwargs=dict( project="consistent-teacher", name="${cfg_name}", config=dict( fold="${fold}", percent="${percent}", work_dirs="${work_dir}", total_step="${runner.max_iters}", ), ), by_epoch=False, ) ], ) Where are you talking about train_cfg 原始邮件 发件人："Xingyi Yang"< ***@***.*** &gt;; 发件时间：2024/1/17 19:54 收件人："Adamdad/ConsistentTeacher"< ***@***.*** &gt;; 抄送人："Code-of-Liujie"< ***@***.*** &gt;;"Comment"< ***@***.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Do you want to try it remotely? 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 18:56 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Please indicate which config you are using. For all configs in this repo, we have model.train_cfg. For example https://github.com/Adamdad/ConsistentTeacher/blob/b49a9ce1c0ae78d13528be0a9a4eff65a627d581/configs/consistent-teacher/consistent_teacher_r50_fpn_coco_180k_10p.py#L52-L58 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***&gt; Sorry, but what do you mean by "try it remotely"? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***&gt;

[Code-of-Liujie]

Remote control is to remotely operate my computer through the sunflower software, I see 原始邮件 发件人："Xingyi Yang"< ***@***.*** &gt;; 发件时间：2024/1/17 19:54 收件人："Adamdad/ConsistentTeacher"< ***@***.*** &gt;; 抄送人："Code-of-Liujie"< ***@***.*** &gt;;"Comment"< ***@***.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Do you want to try it remotely? 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 18:56 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Please indicate which config you are using. For all configs in this repo, we have model.train_cfg. For example https://github.com/Adamdad/ConsistentTeacher/blob/b49a9ce1c0ae78d13528be0a9a4eff65a627d581/configs/consistent-teacher/consistent_teacher_r50_fpn_coco_180k_10p.py#L52-L58 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***&gt; Sorry, but what do you mean by "try it remotely"? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***&gt;

[Adamdad]

base = './consistent_teacher_r50_fpn_coco_180k_10p.py' fold = 1 percent = 10 data = dict( train=dict( sup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}.json", ), unsup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}-unlabeled.json", ), ), ) log_config = dict( delete=True, interval=50, hooks=[ dict(type="TextLoggerHook", by_epoch=False), dict( type="WandbLoggerHook", init_kwargs=dict( project="consistent-teacher", name="${cfg_name}", config=dict( fold="${fold}", percent="${percent}", work_dirs="${work_dir}", total_step="${runner.max_iters}", ), ), by_epoch=False, ) ], ) Where are you talking about train_cfg 原始邮件 发件人："Xingyi Yang"< @.*** >; 发件时间：2024/1/17 19:54 收件人："Adamdad/ConsistentTeacher"< @.*** >; 抄送人："Code-of-Liujie"< @.*** >;"Comment"< @.*** >; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Do you want to try it remotely? 原始邮件 发件人："Xingyi Yang"< @.*** >; 发件时间：2024/1/17 18:56 收件人："Adamdad/ConsistentTeacher"< @.*** >; 抄送人："Code-of-Liujie"< @.*** >;"Comment"< @.*** >; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Please indicate which config you are using. For all configs in this repo, we have model.train_cfg. For example

ConsistentTeacher/configs/consistent-teacher/consistent_teacher_r50_fpn_coco_180k_10p.py

Lines 52 to 58 in b49a9ce

	train_cfg=dict(
	assigner=dict(type='DynamicSoftLabelAssigner', topk=13, iou_factor=2.0),
	alpha=1,
	beta=6,
	allowed_border=-1,
	pos_weight=-1,
	debug=False),

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> Sorry, but what do you mean by "try it remotely"? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

In the first lime, we defaultly import hyper-parameters from another base config

_base_ = './consistent_teacher_r50_fpn_coco_180k_10p.py'

the train_cfg is in the base config.

Sorry, but I cannot not assist you through remote control.

[Code-of-Liujie]

Here this mmdetion is the latest version, but the mmdet in the environment ==2.28.1 is not caused by this reason. 原始邮件 发件人："Xingyi Yang"< ***@***.*** &gt;; 发件时间：2024/1/17 20:09 收件人："Adamdad/ConsistentTeacher"< ***@***.*** &gt;; 抄送人："Code-of-Liujie"< ***@***.*** &gt;;"Comment"< ***@***.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) base = './consistent_teacher_r50_fpn_coco_180k_10p.py' fold = 1 percent = 10 data = dict( train=dict( sup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}.json", ), unsup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}-unlabeled.json", ), ), ) log_config = dict( delete=True, interval=50, hooks=[ dict(type="TextLoggerHook", by_epoch=False), dict( type="WandbLoggerHook", init_kwargs=dict( project="consistent-teacher", name="${cfg_name}", config=dict( fold="${fold}", percent="${percent}", work_dirs="${work_dir}", total_step="${runner.max_iters}", ), ), by_epoch=False, ) ], ) Where are you talking about train_cfg 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 19:54 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Do you want to try it remotely? 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 18:56 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Please indicate which config you are using. For all configs in this repo, we have model.train_cfg. For example https://github.com/Adamdad/ConsistentTeacher/blob/b49a9ce1c0ae78d13528be0a9a4eff65a627d581/configs/consistent-teacher/consistent_teacher_r50_fpn_coco_180k_10p.py#L52-L58 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.&gt; Sorry, but what do you mean by "try it remotely"? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.&gt; In the first lime, we defaultly import hyper-parameters from another base config _base_ = './consistent_teacher_r50_fpn_coco_180k_10p.py' the train_cfg is in the base config. Sorry, but I cannot not assist you through remote control. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***&gt;

[Adamdad]

Here this mmdetion is the latest version, but the mmdet in the environment ==2.28.1 is not caused by this reason. 原始邮件 发件人："Xingyi Yang"< @.*** >; 发件时间：2024/1/17 20:09 收件人："Adamdad/ConsistentTeacher"< @.*** >; 抄送人："Code-of-Liujie"< @.*** >;"Comment"< @.*** >; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) base = './consistent_teacher_r50_fpn_coco_180k_10p.py' fold = 1 percent = 10 data = dict( train=dict( sup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}.json", ), unsup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}-unlabeled.json", ), ), ) log_config = dict( delete=True, interval=50, hooks=[ dict(type="TextLoggerHook", by_epoch=False), dict( type="WandbLoggerHook", init_kwargs=dict( project="consistent-teacher", name="${cfg_name}", config=dict( fold="${fold}", percent="${percent}", work_dirs="${work_dir}", total_step="${runner.max_iters}", ), ), by_epoch=False, ) ], ) Where are you talking about train_cfg 原始邮件 发件人："Xingyi Yang"< @.*** >; 发件时间：2024/1/17 19:54 收件人："Adamdad/ConsistentTeacher"< @.*** >; 抄送人："Code-of-Liujie"< @.*** >;"Comment"< @.*** >; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Do you want to try it remotely? 原始邮件 发件人："Xingyi Yang"< @.*** >; 发件时间：2024/1/17 18:56 收件人："Adamdad/ConsistentTeacher"< @.*** >; 抄送人："Code-of-Liujie"< @.*** >;"Comment"< @.*** >; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Please indicate which config you are using. For all configs in this repo, we have model.train_cfg. For example

ConsistentTeacher/configs/consistent-teacher/consistent_teacher_r50_fpn_coco_180k_10p.py

Lines 52 to 58 in b49a9ce

	train_cfg=dict(
	assigner=dict(type='DynamicSoftLabelAssigner', topk=13, iou_factor=2.0),
	alpha=1,
	beta=6,
	allowed_border=-1,
	pos_weight=-1,
	debug=False),

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> Sorry, but what do you mean by "try it remotely"? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> In the first lime, we defaultly import hyper-parameters from another base config base = './consistent_teacher_r50_fpn_coco_180k_10p.py' the train_cfg is in the base config. Sorry, but I cannot not assist you through remote control. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

I would suggest to down-grade the mmdetection version of base config as well.

[Code-of-Liujie]

The lower version of mmdecion doesn't. Is it possible to do without mmdetection in this directory? 原始邮件 发件人："Xingyi Yang"< ***@***.*** &gt;; 发件时间：2024/1/17 20:18 收件人："Adamdad/ConsistentTeacher"< ***@***.*** &gt;; 抄送人："Code-of-Liujie"< ***@***.*** &gt;;"Comment"< ***@***.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Here this mmdetion is the latest version, but the mmdet in the environment ==2.28.1 is not caused by this reason. 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 20:09 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) base = './consistent_teacher_r50_fpn_coco_180k_10p.py' fold = 1 percent = 10 data = dict( train=dict( sup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}.json", ), unsup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}-unlabeled.json", ), ), ) log_config = dict( delete=True, interval=50, hooks=[ dict(type="TextLoggerHook", by_epoch=False), dict( type="WandbLoggerHook", init_kwargs=dict( project="consistent-teacher", name="${cfg_name}", config=dict( fold="${fold}", percent="${percent}", work_dirs="${work_dir}", total_step="${runner.max_iters}", ), ), by_epoch=False, ) ], ) Where are you talking about train_cfg 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 19:54 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Do you want to try it remotely? 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 18:56 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Please indicate which config you are using. For all configs in this repo, we have model.train_cfg. For example https://github.com/Adamdad/ConsistentTeacher/blob/b49a9ce1c0ae78d13528be0a9a4eff65a627d581/configs/consistent-teacher/consistent_teacher_r50_fpn_coco_180k_10p.py#L52-L58 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.&gt; Sorry, but what do you mean by "try it remotely"? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.&gt; In the first lime, we defaultly import hyper-parameters from another base config base = './consistent_teacher_r50_fpn_coco_180k_10p.py' the train_cfg is in the base config. Sorry, but I cannot not assist you through remote control. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***&gt; I would suggest to down-grade the mmdetection version of base config as well. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***&gt;

[Adamdad]

The lower version of mmdecion doesn't. Is it possible to do without mmdetection in this directory? 原始邮件 发件人："Xingyi Yang"< @.*** >; 发件时间：2024/1/17 20:18 收件人："Adamdad/ConsistentTeacher"< @.*** >; 抄送人："Code-of-Liujie"< @.*** >;"Comment"< @.*** >; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Here this mmdetion is the latest version, but the mmdet in the environment ==2.28.1 is not caused by this reason. 原始邮件 发件人："Xingyi Yang"< @.*** >; 发件时间：2024/1/17 20:09 收件人："Adamdad/ConsistentTeacher"< @.*** >; 抄送人："Code-of-Liujie"< @.*** >;"Comment"< @.*** >; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) base = './consistent_teacher_r50_fpn_coco_180k_10p.py' fold = 1 percent = 10 data = dict( train=dict( sup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}.json", ), unsup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}-unlabeled.json", ), ), ) log_config = dict( delete=True, interval=50, hooks=[ dict(type="TextLoggerHook", by_epoch=False), dict( type="WandbLoggerHook", init_kwargs=dict( project="consistent-teacher", name="${cfg_name}", config=dict( fold="${fold}", percent="${percent}", work_dirs="${work_dir}", total_step="${runner.max_iters}", ), ), by_epoch=False, ) ], ) Where are you talking about train_cfg 原始邮件 发件人："Xingyi Yang"< @.*** >; 发件时间：2024/1/17 19:54 收件人："Adamdad/ConsistentTeacher"< @.*** >; 抄送人："Code-of-Liujie"< @.*** >;"Comment"< @.*** >; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Do you want to try it remotely? 原始邮件 发件人："Xingyi Yang"< @.*** >; 发件时间：2024/1/17 18:56 收件人："Adamdad/ConsistentTeacher"< @.*** >; 抄送人："Code-of-Liujie"< @.*** >;"Comment"< @.*** >; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Please indicate which config you are using. For all configs in this repo, we have model.train_cfg. For example

ConsistentTeacher/configs/consistent-teacher/consistent_teacher_r50_fpn_coco_180k_10p.py

Lines 52 to 58 in b49a9ce

	train_cfg=dict(
	assigner=dict(type='DynamicSoftLabelAssigner', topk=13, iou_factor=2.0),
	alpha=1,
	beta=6,
	allowed_border=-1,
	pos_weight=-1,
	debug=False),

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> Sorry, but what do you mean by "try it remotely"? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> In the first lime, we defaultly import hyper-parameters from another base config base = './consistent_teacher_r50_fpn_coco_180k_10p.py' the train_cfg is in the base config. Sorry, but I cannot not assist you through remote control. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> I would suggest to down-grade the mmdetection version of base config as well. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

You can also copy the older version config of mmdet into local folder, and change the base config path in your current config.

[Code-of-Liujie]

&nbsp; &nbsp; wandb_login._login(anonymous=anonymous, force=force, _disable_warning=True) &nbsp; File "D:\Anaconda3\envs\teacher\lib\site-packages\wandb\sdk\wandb_login.py", line 238, in _login &nbsp; &nbsp; wlogin.prompt_api_key() &nbsp; File "D:\Anaconda3\envs\teacher\lib\site-packages\wandb\sdk\wandb_login.py", line 174, in prompt_api_key &nbsp; &nbsp; raise UsageError("api_key not configured (no-tty). call " + directive) wandb.errors.UsageError: api_key not configured (no-tty). call wandb.login(key=[your_api_key]) 原始邮件 发件人："Xingyi Yang"< ***@***.*** &gt;; 发件时间：2024/1/17 20:24 收件人："Adamdad/ConsistentTeacher"< ***@***.*** &gt;; 抄送人："Code-of-Liujie"< ***@***.*** &gt;;"Comment"< ***@***.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) The lower version of mmdecion doesn't. Is it possible to do without mmdetection in this directory? 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 20:18 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Here this mmdetion is the latest version, but the mmdet in the environment ==2.28.1 is not caused by this reason. 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 20:09 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) base = './consistent_teacher_r50_fpn_coco_180k_10p.py' fold = 1 percent = 10 data = dict( train=dict( sup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}.json", ), unsup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}-unlabeled.json", ), ), ) log_config = dict( delete=True, interval=50, hooks=[ dict(type="TextLoggerHook", by_epoch=False), dict( type="WandbLoggerHook", init_kwargs=dict( project="consistent-teacher", name="${cfg_name}", config=dict( fold="${fold}", percent="${percent}", work_dirs="${work_dir}", total_step="${runner.max_iters}", ), ), by_epoch=False, ) ], ) Where are you talking about train_cfg 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 19:54 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Do you want to try it remotely? 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 18:56 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Please indicate which config you are using. For all configs in this repo, we have model.train_cfg. For example https://github.com/Adamdad/ConsistentTeacher/blob/b49a9ce1c0ae78d13528be0a9a4eff65a627d581/configs/consistent-teacher/consistent_teacher_r50_fpn_coco_180k_10p.py#L52-L58 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.&gt; Sorry, but what do you mean by "try it remotely"? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.&gt; In the first lime, we defaultly import hyper-parameters from another base config base = './consistent_teacher_r50_fpn_coco_180k_10p.py' the train_cfg is in the base config. Sorry, but I cannot not assist you through remote control. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.&gt; I would suggest to down-grade the mmdetection version of base config as well. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.&gt; You can also copy the older version config of mmdet into local folder, and change the base config path in your current config. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***&gt;

[Adamdad]

    wandb_login._login(anonymous=anonymous, force=force, _disable_warning=True)   File "D:\Anaconda3\envs\teacher\lib\site-packages\wandb\sdk\wandb_login.py", line 238, in _login     wlogin.prompt_api_key()   File "D:\Anaconda3\envs\teacher\lib\site-packages\wandb\sdk\wandb_login.py", line 174, in prompt_api_key     raise UsageError("api_key not configured (no-tty). call " + directive) wandb.errors.UsageError: api_key not configured (no-tty). call wandb.login(key=[your_api_key]) 原始邮件 发件人："Xingyi Yang"< @.*** >; 发件时间：2024/1/17 20:24 收件人："Adamdad/ConsistentTeacher"< @.*** >; 抄送人："Code-of-Liujie"< @.*** >;"Comment"< @.*** >; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) The lower version of mmdecion doesn't. Is it possible to do without mmdetection in this directory? 原始邮件 发件人："Xingyi Yang"< @.*** >; 发件时间：2024/1/17 20:18 收件人："Adamdad/ConsistentTeacher"< @.*** >; 抄送人："Code-of-Liujie"< @.*** >;"Comment"< @.*** >; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Here this mmdetion is the latest version, but the mmdet in the environment ==2.28.1 is not caused by this reason. 原始邮件 发件人："Xingyi Yang"< @.*** >; 发件时间：2024/1/17 20:09 收件人："Adamdad/ConsistentTeacher"< @.*** >; 抄送人："Code-of-Liujie"< @.*** >;"Comment"< @.*** >; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) base = './consistent_teacher_r50_fpn_coco_180k_10p.py' fold = 1 percent = 10 data = dict( train=dict( sup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}.json", ), unsup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}-unlabeled.json", ), ), ) log_config = dict( delete=True, interval=50, hooks=[ dict(type="TextLoggerHook", by_epoch=False), dict( type="WandbLoggerHook", init_kwargs=dict( project="consistent-teacher", name="${cfg_name}", config=dict( fold="${fold}", percent="${percent}", work_dirs="${work_dir}", total_step="${runner.max_iters}", ), ), by_epoch=False, ) ], ) Where are you talking about train_cfg 原始邮件 发件人："Xingyi Yang"< @.*** >; 发件时间：2024/1/17 19:54 收件人："Adamdad/ConsistentTeacher"< @.*** >; 抄送人："Code-of-Liujie"< @.*** >;"Comment"< @.*** >; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Do you want to try it remotely? 原始邮件 发件人："Xingyi Yang"< @.*** >; 发件时间：2024/1/17 18:56 收件人："Adamdad/ConsistentTeacher"< @.*** >; 抄送人："Code-of-Liujie"< @.*** >;"Comment"< @.*** >; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Please indicate which config you are using. For all configs in this repo, we have model.train_cfg. For example

ConsistentTeacher/configs/consistent-teacher/consistent_teacher_r50_fpn_coco_180k_10p.py

Lines 52 to 58 in b49a9ce

	train_cfg=dict(
	assigner=dict(type='DynamicSoftLabelAssigner', topk=13, iou_factor=2.0),
	alpha=1,
	beta=6,
	allowed_border=-1,
	pos_weight=-1,
	debug=False),

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> Sorry, but what do you mean by "try it remotely"? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> In the first lime, we defaultly import hyper-parameters from another base config base = './consistent_teacher_r50_fpn_coco_180k_10p.py' the train_cfg is in the base config. Sorry, but I cannot not assist you through remote control. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> I would suggest to down-grade the mmdetection version of base config as well. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> You can also copy the older version config of mmdet into local folder, and change the base config path in your current config. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

please register on wandb and put your username and api-key in the code.

[Code-of-Liujie]

原始邮件 发件人："Xingyi Yang"< ***@***.*** &gt;; 发件时间：2024/1/17 20:24 收件人："Adamdad/ConsistentTeacher"< ***@***.*** &gt;; 抄送人："Code-of-Liujie"< ***@***.*** &gt;;"Comment"< ***@***.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) The lower version of mmdecion doesn't. Is it possible to do without mmdetection in this directory? 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 20:18 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Here this mmdetion is the latest version, but the mmdet in the environment ==2.28.1 is not caused by this reason. 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 20:09 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) base = './consistent_teacher_r50_fpn_coco_180k_10p.py' fold = 1 percent = 10 data = dict( train=dict( sup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}.json", ), unsup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}-unlabeled.json", ), ), ) log_config = dict( delete=True, interval=50, hooks=[ dict(type="TextLoggerHook", by_epoch=False), dict( type="WandbLoggerHook", init_kwargs=dict( project="consistent-teacher", name="${cfg_name}", config=dict( fold="${fold}", percent="${percent}", work_dirs="${work_dir}", total_step="${runner.max_iters}", ), ), by_epoch=False, ) ], ) Where are you talking about train_cfg 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 19:54 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Do you want to try it remotely? 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 18:56 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Please indicate which config you are using. For all configs in this repo, we have model.train_cfg. For example https://github.com/Adamdad/ConsistentTeacher/blob/b49a9ce1c0ae78d13528be0a9a4eff65a627d581/configs/consistent-teacher/consistent_teacher_r50_fpn_coco_180k_10p.py#L52-L58 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.&gt; Sorry, but what do you mean by "try it remotely"? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.&gt; In the first lime, we defaultly import hyper-parameters from another base config base = './consistent_teacher_r50_fpn_coco_180k_10p.py' the train_cfg is in the base config. Sorry, but I cannot not assist you through remote control. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.&gt; I would suggest to down-grade the mmdetection version of base config as well. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.&gt; You can also copy the older version config of mmdet into local folder, and change the base config path in your current config. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***&gt;

[Code-of-Liujie]

原始邮件 发件人："Xingyi Yang"< ***@***.*** &gt;; 发件时间：2024/1/17 21:19 收件人："Adamdad/ConsistentTeacher"< ***@***.*** &gt;; 抄送人："Code-of-Liujie"< ***@***.*** &gt;;"Comment"< ***@***.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) &nbsp; &nbsp; wandb_login._login(anonymous=anonymous, force=force, _disable_warning=True) &nbsp; File "D:\Anaconda3\envs\teacher\lib\site-packages\wandb\sdk\wandb_login.py", line 238, in _login &nbsp; &nbsp; wlogin.prompt_api_key() &nbsp; File "D:\Anaconda3\envs\teacher\lib\site-packages\wandb\sdk\wandb_login.py", line 174, in prompt_api_key &nbsp; &nbsp; raise UsageError("api_key not configured (no-tty). call " + directive) wandb.errors.UsageError: api_key not configured (no-tty). call wandb.login(key=[your_api_key]) 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 20:24 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) The lower version of mmdecion doesn't. Is it possible to do without mmdetection in this directory? 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 20:18 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Here this mmdetion is the latest version, but the mmdet in the environment ==2.28.1 is not caused by this reason. 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 20:09 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) base = './consistent_teacher_r50_fpn_coco_180k_10p.py' fold = 1 percent = 10 data = dict( train=dict( sup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}.json", ), unsup=dict( ann_file="data/coco_semi/semi_supervised/instances_train2017.${fold}@${percent}-unlabeled.json", ), ), ) log_config = dict( delete=True, interval=50, hooks=[ dict(type="TextLoggerHook", by_epoch=False), dict( type="WandbLoggerHook", init_kwargs=dict( project="consistent-teacher", name="${cfg_name}", config=dict( fold="${fold}", percent="${percent}", work_dirs="${work_dir}", total_step="${runner.max_iters}", ), ), by_epoch=False, ) ], ) Where are you talking about train_cfg 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 19:54 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Do you want to try it remotely? 原始邮件 发件人："Xingyi Yang"< @.*** &gt;; 发件时间：2024/1/17 18:56 收件人："Adamdad/ConsistentTeacher"< @.*** &gt;; 抄送人："Code-of-Liujie"< @.*** &gt;;"Comment"< @.*** &gt;; 主题：Re: [Adamdad/ConsistentTeacher] Multi card parallel (Issue #51) Please indicate which config you are using. For all configs in this repo, we have model.train_cfg. For example https://github.com/Adamdad/ConsistentTeacher/blob/b49a9ce1c0ae78d13528be0a9a4eff65a627d581/configs/consistent-teacher/consistent_teacher_r50_fpn_coco_180k_10p.py#L52-L58 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.&gt; Sorry, but what do you mean by "try it remotely"? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.&gt; In the first lime, we defaultly import hyper-parameters from another base config base = './consistent_teacher_r50_fpn_coco_180k_10p.py' the train_cfg is in the base config. Sorry, but I cannot not assist you through remote control. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.&gt; I would suggest to down-grade the mmdetection version of base config as well. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.&gt; You can also copy the older version config of mmdet into local folder, and change the base config path in your current config. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***&gt; please register on wandb and put your username and api-key in the code. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***&gt;