search

NVlabs / Deep_Object_Pose

Deep Object Pose Estimation (DOPE) – ROS inference (CoRL 2018)
Other
1.03k stars 287 forks source link

Range of detection #374

Open AyeshaHabib opened 4 months ago

AyeshaHabib commented 4 months ago

Hello I have been working on the DOPE for the orange color module using webcam and realsense camera but the issue is for the range of detection. As soon as object gets in range of 50cm of camera, it stops detecting and only detects for the certain range. I tried calibrating camera extrinsic to work around it but it didn't worked either and still give the same result. The scale I used in configuration script is the dimension of actual object and its 3D cad model. I have attached the picture from NVISII data generated for reference. KIndly let me know how to improve range of detection 00057

AyeshaHabib commented 4 months ago

results

RenanMoreiraPinto commented 4 months ago

hey you used ros1? do you have a docker version you using? thanks!

TontonTremblay commented 4 months ago

I think you will need renders in your training data where the object appears closer.

AyeshaHabib commented 4 months ago

hey you used ros1? do you have a docker version you using? thanks! I used ros1 installed in my workstation already, haven't tried using docker version though.

AyeshaHabib commented 4 months ago

@TontonTremblay Can you please eloborate in it more, will be of great help. What exactly should I configure in single_video_pybullet for object to appear closer to camera in data generated using NVISII.