segmentation_labels is empty dict

[rttariverdi67]

Hi all, I run tutorial_3d_semseg notebook,

(range_images, camera_projections, segmentation_labels,
 range_image_top_pose) = frame_utils.parse_range_image_and_camera_projection( frame)

here the dict is empty!

thanks for your help in advance :)

[LinaShanghaitech]

Hi all, I run tutorial_3d_semseg notebook,

(range_images, camera_projections, segmentation_labels,
 range_image_top_pose) = frame_utils.parse_range_image_and_camera_projection( frame)

here the dict is empty!

thanks for your help in advance :)

I also have this problem and keypoints annotation is also empty list from proto

[jialeli1]

I also encountered something similar. In a tfrecord file, it seems that only a few frames is annotated with the point-wise semantic segmentation label. Some checks are in this issue.

[cardwing]

@jialeli1 @LinaShanghaitech @rttariverdi67 Hi, all. I wonder if you have successfully submitted the model predictions on val/test set to the online server?

[LinaShanghaitech]

@cardwing , sorry, I just use this dataset and not submit the predictions.

[charlesq34]

@rttariverdi67 Have you used the V1.3 Waymo Open Dataset or the old version? The 3d semseg labels only exist in V1.3.

Regarding the annotated frames, we only have 3d semseg labels on a subset of the frames (2Hz/every 5 frames for the frames in the mid-part of the sequence).

[SizheWei]

@charlesq34 Hi, is there any convenient method for me to find out which tfrecords in the dataset v1.3 contain 3D annotations more quickly? Or I need to iterate over all frames of all tfrecords? Thanks!

@rttariverdi67 Have you used the V1.3 Waymo Open Dataset or the old version? The 3d semseg labels only exist in V1.3.

Regarding the annotated frames, we only have 3d semseg labels on a subset of the frames (2Hz/every 5 frames for the frames in the mid-part of the sequence).

[charlesq34]

@SizheWei I believe we have 3d semseg labels on every run segment/tfrecord file. To find out which frames have labels, yes, you'll need to iterate over the frames in each tfrecord file.

[fabianwindbacher]

Can I generally expect, that if an image of a frame has keypoint annotations, that all visible humans are annotated?