data

[herochen7372]

Can you tell me how I can match the 3D coordinates in the json file with the image in human36M?

[Zhuyue0324]

We don't use this image path by default, but using our data_preparation.py script to compile image from video.

However, you can check the closed issue #12 for how to convert between our paths to your format.

[Zhuyue0324]

[herochen7372]

Sorry，I mean this IDEA-Research/OSX#108

[Zhuyue0324]

This is the original site where I get the image data from. I don't know what is the act 16 (and 17 if there is one in your list)

[herochen7372]

Ok，thank you！

[Zhuyue0324]

In worst case, you can just ignore the act 16 and try matching annotations of the 15 actions we provide. (Though I am not sure if your act 1-15 are exactly the actions we provide, you need check by yourself)

[herochen7372]

Ok，thank you，i will try matching 02-16 to 15 actions you provide.

[herochen7372]

I compare annotation of osx' Human36M and your H3WD，and find that they have different frames. Your video frame annotations are not continuous, e.g. there is no S1/Images/Directions.60457274/frame_0001.jpg.

[Zhuyue0324]

No, they are not continuous. First I only compile 1 from each 5 frames from video. Then during the quality assessment step we choose to form only 100k sample triplets out of all samples we compiled with the smallest errors according to our 4-corner error metric.

[herochen7372]

Ok，thank you！

[Z-mingyu]

No, they are not continuous. First I only compile 1 from each 5 frames from video. Then during the quality assessment step we choose to form only 100k sample triplets out of all samples we compiled with the smallest errors according to our 4-corner error metric.

hello ,thanks for your dataset. how can I get the image you used in the dataset

[Zhuyue0324]

I first downloaded the video data from their original site http://vision.imar.ro/human3.6m/description.php
Then I use the code in data_preparation.py to convert mp4 into images

[Z-mingyu]

I first downloaded the video data from their original site http://vision.imar.ro/human3.6m/description.php Then I use the code in data_preparation.py to convert mp4 into images

I found the frame_id in the dataset, so I can get the image you selected