# 3d ans 3d object detection questions **URL:** \ **Category:** ROS Perception in 5 Days\ **Created:** [February 7, 2020, 2:15am UTC](https://get-help.theconstruct.ai/t/3d-ans-3d-object-detection-questions/2211 "2020-02-07T02:15:46Z")\ **Posts on this page:** 3\ **Page:** 1
**Author:** ![fgenestcvdl](https://get-help.theconstruct.ai/user_avatar/get-help.theconstruct.ai/fgenestcvdl/32/9348_2.png) [@fgenestcvdl](https://get-help.theconstruct.ai/u/fgenestcvdl)\ **Post date:** [February 7, 2020, 2:15am UTC](https://get-help.theconstruct.ai/t/3d-ans-3d-object-detection-questions/2211/1 "2020-02-07T02:15:46Z")
In order to have to have 3D detection do we need to do both 2D and 3D image training ? When we have many object and many images of all of them let say 10 coke can images 20 beer can images Is there a way to tell the detector that i have 2 different objects not 20 ? Is there a way to access the coordinates of the detected objects ( like we have the table\_array ) ? So we can choose to pickup the coke can or the bear can ( as an exemple ) ---
**Author:** ![duckfrost2](https://get-help.theconstruct.ai/user_avatar/get-help.theconstruct.ai/duckfrost2/32/9365_2.png) [@duckfrost2](https://get-help.theconstruct.ai/u/duckfrost2)\ **Post date:** [February 10, 2020, 12:47pm UTC](https://get-help.theconstruct.ai/t/3d-ans-3d-object-detection-questions/2211/2 "2020-02-10T12:47:08Z")
The object detection in this curse is really simple so what you are saing woulndt be done easily. I would reccomend you having a look to the manipulation course, which dwelved more in this matter. Also we have a course that uses deeplearning with random environments to position a certain object. [Manipulation Couse](https://www.robotigniteacademy.com/en/course/ros-manipulation-in-5-days/details/) [Random Env Object detection](https://www.robotigniteacademy.com/en/course/deep-learning-with-domain-randomization/details/) ---
**Author:** ![fgenestcvdl](https://get-help.theconstruct.ai/user_avatar/get-help.theconstruct.ai/fgenestcvdl/32/9348_2.png) [@fgenestcvdl](https://get-help.theconstruct.ai/u/fgenestcvdl)\ **Post date:** [February 12, 2020, 3:07pm UTC](https://get-help.theconstruct.ai/t/3d-ans-3d-object-detection-questions/2211/3 "2020-02-12T15:07:44Z")
Thanks for the advices about these other modules but i would like to be able to use the results of the detections obtained by this module. I have problems interpreting the **find\_object\_2d/ObjectsStamped** message. From rosmessage info i get a description that is not clear for me see below: **rosmsg info find\_object\_2d/ObjectsStamped** std\_msgs/Header header uint32 seq time stamp string frame\_id std\_msgs/Float32MultiArray objects std\_msgs/MultiArrayLayout layout std\_msgs/MultiArrayDimension dim string label uint32 size uint32 stride uint32 data\_offset float32 data Here is an exemple of the topic echo when i detect the table in front of the robot ( and below the understanding that i have of it) ser:~$ **rostopic echo /objectsStamped** header: seq: 894 stamp: secs: 1172 nsecs: 134000000 frame\_id: “head\_camera\_rgb\_optical\_frame” objects: layout: dim: data\_offset: 0 data: [6.0, 382.0, 146.0, 0.9914313554763794, -0.008840509690344334, -3.84211161872372e-05, 0.018584784120321274, 1.0304908752441406, 9.121886250795797e-05, 68.8837890625, 265.6440734863281, 1.0] **Interpretation** 6 Object label (it is the number the image file in the directory saved\_pictures2d ex: 6.png ) 382 Object x size in pixels ? 146 Object y size in pixels ? 0.9914313554763794, -0.008840509690344334, -3.84211161872372e-05 **( looks like x1,y1,z1 ?)** 0.018584784120321274, 1.0304908752441406, 9.121886250795797e-05 **( looks like x2,y2,z2 ?)** 9.121886250795797e-05, 68.8837890625, 265.6440734863281, 1.0 **( looks like x3,y3,z3 ?)** I would have expected the object pose at it center but it is looking to be something else?