How to calculate the result [email protected]? #30
Comments
|
Dear Le,
The PCK is computed as average for only the visible (=annotated) key points.
The zebras are not all toward right, as you can see in the paper, Fig. 6.
I hope this answers your questions,
Best,
Silvia
… On 19. Nov 2021, at 18:21, Drow999 ***@***.***> wrote:
Hello Docter Zuffi,
I read the paper and want to test your network on another testset, but I'm not sure how you process the result, because in your annotation there are many invisible points. Did you ignore the invisible points when you compare them with the estimation points? Also, how did you calculate the total average of the PCK. Is it a simple average of all bodyparts? I think the average will be greatly lowered because there are some bodyparts always invisible (like left shoulder, you made all zebra towards right). By the way, the PCK you chose is the PCK head? Thank you!
Best,
Le
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub <#30>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABNYVFOKONR7CS4Z553LRWDUM2BTLANCNFSM5IMUFNDA>.
Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>.
|
|
I understand the first question now. And, I mean in the test dataset (100 images) you used, the zebras are all towards the right, aren't they? And the PCK you used here is the PCKh? In Benjamin Biggs work which is similar to yours, he used the size of the image as the factor but not the head. It made me quite confused. |
|
The test images are mirrored.
I do not have keypoints that I treat in different way.
Best,
Silvia
… On 19. Nov 2021, at 18:44, Drow999 ***@***.***> wrote:
I understand the first question now. And, I mean in the test dataset (100 images) you used, the zebras are all towards the right, aren't they? And the PCK you used here is the PCKh? In Benjamin Biggs work which is similar to yours, he used the size of the image as the factor but not the head. It made me quite confused.
—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub <#30 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABNYVFIIF7NKMAWNJCQLNXTUM2EGHANCNFSM5IMUFNDA>.
Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>.
|
|
Thank you! It's clear. |
|
The image size.
… On 19. Nov 2021, at 20:19, Drow999 ***@***.***> wrote:
Thank you! It's clear.
I'm reading the code eval.py, and I have a question about the way you calculate the error. Why did you use kp/256 - kp_pred/256 < alpha. what does the 256 mean here.
<https://user-images.githubusercontent.com/77601118/142678654-001e2117-8522-4b24-b2cb-306b36901339.png>
Best,
Le
—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub <#30 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABNYVFOXUTYIIXIH2CK67S3UM2PMDANCNFSM5IMUFNDA>.
Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>.
|
|
so you didn't use zebra's head but the image size as the scalar, is that right? This part is really confusing, could you tell me why you choose to resize the image before you predict the keypoints? Will you have a better result on a smaller image? Best, |
|
The image is resized to 256 before predicting the keypoints because the network is trained on 256x256 images.
… On 20. Nov 2021, at 08:41, Drow999 ***@***.***> wrote:
so you didn't use zebra's head but the image size as the scalar, is that right? This part is really confusing, could you tell me why you choose to resize the image before you predict the keypoints? Will you have a better result on a smaller image?
Best,
Le
—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub <#30 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABNYVFIBEORVZFM6I3MMDGDUM5GMDANCNFSM5IMUFNDA>.
Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>.
|
|
Thank you so much for your explanation! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Hello Docter Zuffi,
I read the paper and want to test your network on another testset, but I'm not sure how you process the result, because in your annotation there are many invisible points. Did you ignore the invisible points when you compare them with the estimation points? Also, how did you calculate the total average of the PCK. Is it a simple average of all bodyparts? I think the average will be greatly lowered because there are some bodyparts always invisible (like left shoulder, you made all zebra towards right). By the way, the PCK you chose is the PCK head? Thank you!
Best,
Le
The text was updated successfully, but these errors were encountered: