Argoverse 3D Detection Competition - submission format ??

[cdahms123]

I've worked with some other AV datasets before and now I'm planning on making a submission to the Argoverse 3D Detection Competition https://eval.ai/web/challenges/challenge-page/725/submission.

I'm unable to find explanation or documentation of the submission format for this competition. Is this available somewhere? Or can somebody explain the format?

I would have expected documentation somewhere saying something to the effect of:

upload a zipped .csv file with your detections in the following format:

frameId0 x0,y0,z0,w0,l0,h0,yaw0 x1,y1,z1,w1,l1,h1,yaw1 x2,y2,z2,w2,l2,h2,yaw2 . . .
frameId1 x0,y0,z0,w0,l0,h0,yaw0 x1,y1,z1,w1,l1,h1,yaw1 x2,y2,z2,w2,l2,h2,yaw2 . . .
frameId2 x0,y0,z0,w0,l0,h0,yaw0 x1,y1,z1,w1,l1,h1,yaw1 x2,y2,z2,w2,l2,h2,yaw2 . . .
.
.
.

or similar, but I can't find anything defining the submission format, please advise.

One additional question, on the leaderboard for this competition https://eval.ai/web/challenges/challenge-page/725/leaderboard/1974 some of the entrants make mention of "Baseline", is a tutorial or similar for how to make a submission to this competition specifically publicly available? The tutorials section https://github.com/argoai/argoverse-api/tree/master/demo_usage seems to contain tutorials for tracking, stereo, and map but not object detection.

[johnwlambert]

Hi @cdahms123, thanks for your interest in Argoverse.

The submission format is the same as the release format for the labels, except a track_id is not required. Otherwise, the detection submission format is identical to the tracking submission format found here: