where is images and keypoints for DATA_FOLDER?

[TomatoBoy90]

hi, your job so great .however,how to download those data?

[keeelho]

you need to put image and keypoint, and use openpose to make keypoint

[tetsu]

Do you know what kind of format for images and keypoints? Can we use PNG images and 2D keypoints in JSON format?

[Anirudh257]

@tetsu Yes, the openpose keypoints is in Openpose json format and the images are in .png format.

[poincarelee]

hi, I would like to know how many images processed by openpose are enough to run in the model. @TomatoBoy90 @Anirudh257 @keeelho @geopavlakos

[Anirudh257]

@poincarelee one image is also fine to run the model.

[whois-byte]

Does Data_FOLDER have an existing file link, and what should the file format below it look like?

[Anirudh257]

@who123sy, it is like:

data_folder/
data_folder/images/
data_folder/images/image_1.jpg
data_folder/keypoints/
data_folder/keypoints/image_openpose_1.jpg

[sparshgarg23]

@Anirudh257 is it mandatory to use openpose,or will any keypoint detection like alphapose/deepcut/mediapipe work?

[Anirudh257]

@sparshgarg23 anything will work. Follow the folder structure of the images and key points.

[sparshgarg23]

Ok,but mediapipe's output consists of detecing 33 landmarks(including full body face,hand and landmark),whereas open pose output consists of detecting 25 points.So wouldn't that be a problem when we are reprojecting the 3d points outputted by the model back into 2d? from mediapipe site The landmark model in MediaPipe Pose predicts the location of 33 pose landmarks from openpose Pose Output Format (BODY_25)

I mean the main goal of SMPL is to reduce the error between the projected 3d points and the 2d points,so wouldn't there be a dimensional mismatch. Thanks for taking the time to get back,hope you can clarify my doubt.

[Anirudh257]

@sparshgarg23 Okay, that is true. Try to use the 25 keypoints corresponding to Openpose only. You may have to do some manual changes.