Sample_result data from nuScenes tracking task

#1

Hello ,
In the nuScenes tracking task, under Results format section there is a description for sample_result table. I am trying to run a tracking MOT on nuscenes data which generated results in sample_result table format. Could you please help me understand how to visualize this result on the nuscenes data. Do you have any helper functions in the nuscenes devkit to visualize the result.
Sample output:
“meta”: {
“use_camera”: true,
“use_lidar”: true,
“use_radar”: false,
“use_map”: false,
“use_external”: false
},
“results”: {
“3e8750f331d7499e9b5123e9eb70f2e2”: [
{
“sample_token”: “3e8750f331d7499e9b5123e9eb70f2e2”,
“translation”: [
588.0003239270841,
1599.0074551885257,
-0.35196263493975266
],
“size”: [
1.9676951169967651,
4.607950687408447,
1.811274528503418
],
“rotation”: [
0.8743069621837772,
0.0,
0.0,
0.48537339840268856
],
“velocity”: [
0.48454040137211885,
0.735596687143289
],
“tracking_id”: “108”,
“tracking_name”: “car”,
“tracking_score”: 0.006666450450817744,
“yaw”: -1.0135803927443716
},

Thanks,
Gayathri.

#2

Hi @Gayathri81, there’s a flag here which you can set: https://github.com/nutonomy/nuscenes-devkit/blob/master/python-sdk/nuscenes/eval/tracking/evaluate.py#L248

#3

Hi @Whye_Kit_Fong, thanks for the quick reply. When I ran evaluate file giving car in render classes field I get some images loaded on to the result folder which I couldn’t understand. It will be great if you can help me understanding the boxes in the result image.
Please find the below image for reference.

image
image.png639×585 41.9 KB

#4

@Gayathri81 if you take a look at the evaluation code here, each image is essentially a frame

And within each frame, if you take a look at the render function here, each colored box refers to a track ID

So across frames, you would want to see the color of each box being consistent (otherwise, there could have been some identity switches for which you can improve your tracking algorithm)

#5

@Whye_Kit_Fong Thanks for the quick clarification.