search

Xianpeng919 / MonoCon

Learning Auxiliary Monocular Contexts Helps Monocular 3D Object Detection (AAAI'22)
146 stars 23 forks source link

About performance on personal data #6

Closed wu-hrww closed 2 years ago

wu-hrww commented 2 years ago

Hello, @Xianpeng919 , I really appreciate your awesome work. I plan to deploy your work on real-vehicle platforms for vehicle tracking, but test performance on my personal data is very strange. All 3D boxes are of low scores. With a threshold of 0.1, nearly no box would be outputted. Several examples are shown as follows： result_1 result_2 result_3 And the camera intrinsics matrix is [[1201.8,-0.935,953.9414],[0.000000,1211.80,636.9038],[0.000000,0.000000,1.000000]]. The raw images are attached below. Looking forward to your help. my_data.zip

Xingyi0220 commented 1 year ago

I think that they are not rectified photos

Xingyi0220 commented 1 year ago

Personally thinking, you should rectify the photos and apply the rectified projection matrix