several questions

[SylvainArd]

Hello I have several questions : 1 : what is the latest version 0 or 1 ? and is the version 2 which will arrive in January improve a lot the program ? 2 : in COCO format a point class (ex : left_eye) have one and exactly one point per object and per class, how to have multiple points per object and per class please ? have you a code to advise to me ? 3 : what are the values to put in sigmas in the file configs/base/datasets/coco.py please for a custom dataset ? 4 : have you a description of the differents config files and their differences please ? 5 : how to change the default model (resnet50, resnet101, etc) of a config file ? 6 : what means " type='upper'" in configs/base/datasets/coco.py please 7 : have I to add lines to these lines of \configs\body\2d_kpt_sview_rgb_img\cid\coco\hrnet_w48_coco_512x512.py file if I put image_size to 1024 : extra=dict( stage1=dict( num_modules=1, num_branches=1, block='BOTTLENECK', num_blocks=(4, ), num_channels=(64, )), stage2=dict( num_modules=1, num_branches=2, block='BASIC', num_blocks=(4, 4), num_channels=(48, 96)), stage3=dict( num_modules=4, num_branches=3, block='BASIC', num_blocks=(4, 4, 4), num_channels=(48, 96, 192)), stage4=dict( num_modules=3, num_branches=4, block='BASIC', num_blocks=(4, 4, 4, 4), num_channels=(48, 96, 192, 384), multiscale_output=True)), and what please ? thank you very much best regards

[ly015]

1. MMPose v0.x (latest version is v0.29.0) and v1.x (latest version is v1.0.0rc1) are maintained in parallel now and the next year. New features will be supported in v1.x first.
2. In a dataset, each sample (object) should have a consistent number of keypoints, and the keypoint should be well-defined. Objects with arbitrary numbers of keypoints or multiple keypoints with an identical definition are unsupported in MMPose.
3. COCO website describes how to obtain sigmas from annotations. A custom dataset of human body keypoint can also use sigmas from COCO.

4-7: MMPose documentation provides tutorials for config and custom dataset, which should be able to cover these questions. And more detailed configurations of specific modules can be found in the API reference. It's highly recommended to browse the documentation first before jumping into coding or experimenting, especially for users new to the OpenMMLab framework.

[SylvainArd]

sigmas are not presents if I do Ctrl+F in the page https://cocodataset.org/#keypoints-eval I don't find the answers of the orther questions or for example for the question 7 what I must put is not clear for me

Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 25 nov. 2022 à 13:38, Yining Li @.***> a écrit :

1. MMPose v0.x (latest version is v0.29.0) and v1.x (latest version is v1.0.0rc1) are maintained in parallel now and the next year. New features will be supported in v1.x first.
2. In a dataset, each sample (object) should have a consistent number of keypoints, and the keypoint should be well-defined. Objects with arbitrary numbers of keypoints or multiple keypoints with an identical definition are unsupported in MMPose.
3. COCO website describes how to obtain sigmas from annotations. A custom dataset of human body keypoint can also use sigmas from COCO.

4-7: MMPose documentation provides tutorials for config https://mmpose.readthedocs.io/en/latest/tutorials/0_config.html and custom dataset https://mmpose.readthedocs.io/en/latest/tutorials/2_new_dataset.html, which should be able to cover these questions. And more detailed configurations of specific modules can be found in the API reference https://mmpose.readthedocs.io/en/latest/api.html. It's highly recommended to browse the documentation first before jumping into coding or experimenting, especially for users new to the OpenMMLab framework.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1840#issuecomment-1327423552, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR4MN246ZINZS47XB4DWKCXLZANCNFSM6AAAAAASLHQMKY . You are receiving this because you authored the thread.Message ID: @.***>

[SylvainArd]

if I understand in the mmpose documentation sigmas are standard deviation of keypoints so if I want the keypoints detected to be at one pixel in mean of the ground truth pixel I must put 1/1024=0.001 ? and what are the links of mmpose 1 and mmpose 2 please ? and are their codes very different ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 25 nov. 2022 à 13:52, Sylvain Ard @.***> a écrit :

sigmas are not presents if I do Ctrl+F in the page https://cocodataset.org/#keypoints-eval I don't find the answers of the orther questions or for example for the question 7 what I must put is not clear for me

Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le ven. 25 nov. 2022 à 13:38, Yining Li @.***> a écrit :

1. MMPose v0.x (latest version is v0.29.0) and v1.x (latest version is v1.0.0rc1) are maintained in parallel now and the next year. New features will be supported in v1.x first.
2. In a dataset, each sample (object) should have a consistent number of keypoints, and the keypoint should be well-defined. Objects with arbitrary numbers of keypoints or multiple keypoints with an identical definition are unsupported in MMPose.
3. COCO website describes how to obtain sigmas from annotations. A custom dataset of human body keypoint can also use sigmas from COCO.

4-7: MMPose documentation provides tutorials for config https://mmpose.readthedocs.io/en/latest/tutorials/0_config.html and custom dataset https://mmpose.readthedocs.io/en/latest/tutorials/2_new_dataset.html, which should be able to cover these questions. And more detailed configurations of specific modules can be found in the API reference https://mmpose.readthedocs.io/en/latest/api.html. It's highly recommended to browse the documentation first before jumping into coding or experimenting, especially for users new to the OpenMMLab framework.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1840#issuecomment-1327423552, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR4MN246ZINZS47XB4DWKCXLZANCNFSM6AAAAAASLHQMKY . You are receiving this because you authored the thread.Message ID: @.***>

[ly015]

Please refer to sections 1.2 and 1.3 of https://cocodataset.org/#keypoints-eval about the description of $\sigma$. Actually, I would suggest leaving the sigmas along and using other metrics.

To change the input image size, image_size and heatmap_size is what you need to modify in the config.

There is no mmpose 2. MMPose 0.x (latest is 0.29.0) can be found at the master branch, and its documentation can be found at https://cocodataset.org/#keypoints-eval. MMPose 1.0 can be found at the 1.x branch, and its documentation can be found at https://mmpose.readthedocs.io/en/1.x/. MMPose 1.0 introduced major refactoring and breaking changes.

[SylvainArd]

If I read https://cocodataset.org/#keypoints-eval I see : σi2=E[di2/s2] where di is the distance between the ground truth points and the predicted points and s is the square root of the object area and E is the esperance, as my objects are on average 80 pixels wide, and I want one pixel error maximum, sigma = 1/80=0.01 Am I right ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le lun. 28 nov. 2022 à 06:06, Yining Li @.***> a écrit :

Please refer to sections 1.2 and 1.3 of https://cocodataset.org/#keypoints-eval about the description of $\sigma$. Actually, I would suggest leaving the sigmas along and using other metrics.

To change the input image size, image_size and heatmap_size is what you need to modify in the config.

There is no mmpose 2. MMPose 0.x (latest is 0.29.0) can be found at the master branch, and its documentation can be found at https://cocodataset.org/#keypoints-eval. MMPose 1.0 can be found at the 1.x branch, and its documentation can be found at https://mmpose.readthedocs.io/en/1.x/. MMPose 1.0 introduced major refactoring and breaking changes.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1840#issuecomment-1328545344, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR2JAM26H7FJI32KFQDWKQ4TVANCNFSM6AAAAAASLHQMKY . You are receiving this because you authored the thread.Message ID: @.***>

[SylvainArd]

up please Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le lun. 28 nov. 2022 à 17:35, Sylvain Ard @.***> a écrit :

If I read https://cocodataset.org/#keypoints-eval I see : σi2=E[di2/s 2] where di is the distance between the ground truth points and the predicted points and s is the square root of the object area and E is the esperance, as my objects are on average 80 pixels wide, and I want one pixel error maximum, sigma = 1/80=0.01 Am I right ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le lun. 28 nov. 2022 à 06:06, Yining Li @.***> a écrit :

Please refer to sections 1.2 and 1.3 of https://cocodataset.org/#keypoints-eval about the description of $\sigma$. Actually, I would suggest leaving the sigmas along and using other metrics.

To change the input image size, image_size and heatmap_size is what you need to modify in the config.

There is no mmpose 2. MMPose 0.x (latest is 0.29.0) can be found at the master branch, and its documentation can be found at https://cocodataset.org/#keypoints-eval. MMPose 1.0 can be found at the 1.x branch, and its documentation can be found at https://mmpose.readthedocs.io/en/1.x/. MMPose 1.0 introduced major refactoring and breaking changes.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1840#issuecomment-1328545344, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR2JAM26H7FJI32KFQDWKQ4TVANCNFSM6AAAAAASLHQMKY . You are receiving this because you authored the thread.Message ID: @.***>

[ShaneCan]

ok

[SylvainArd]

So I’m right ?

Envoyé de mon iPhone

Le 30 nov. 2022 à 12:48, ShaneCan @.***> a écrit :

 ok

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.

[ly015]

Sigmas indicate the difference among multiple annotations (from different subjects) of the same keypoint. So you need to collect multiple annotations for each object in your dataset to calculate sigmas.

[SylvainArd]

ok thank you and for my another question of prediction performance please ? Sylvain Ard 0549507724 0778380991 @.*** http://sylvain-ard.fr Entreprise individuelle SIRET : 80079243400022 Appt 26 Bât A Résidence Le Patio 83 rue de la Bugellerie 86000 Poitiers

Le jeu. 1 déc. 2022 à 03:56, Yining Li @.***> a écrit :

Sigmas indicate the diversity among multiple annotations (from different subjects) of the same keypoint. So you need to collect multiple annotations for each object in your dataset to calculate sigmas.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/mmpose/issues/1840#issuecomment-1333086327, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEWZCR5IELGHW4UP6DWZXDLWLAHVVANCNFSM6AAAAAASLHQMKY . You are receiving this because you authored the thread.Message ID: @.***>