[Dataset collection] new Dataset for RoboCup2023

[kvnptl]

• 7 basic object set (existing)
• 9 advanced object set (new)

Total: 16 objects

[kvnptl]

Basic object set (7 objects)

[kvnptl]

Advanced object set (9 objects)

[kvnptl]

• Please find the sheets for dataset collection (set1 and set2) HERE, BIB-CLOUD
• Dataset collection id allotment HERE

Let me know, if you have any doubts

[gokulkchenchani]

• 0 - 154 dataset collection from set 2 completed and updated HERE.
• Dataset collected on different surfaces and on RTT and shelf.

[kvnptl]

Guidelines for data collection:

1. Keep the camera on a tripod

   • Maintain the same height and angle as Youbot's front camera

2. Use rosrun image_view image_view image:=<image topic> to see camera feed

   • Use right-click to save images (NOTE: saved image names start from frame0000)
   • Rename the images according the id(s) given in the sheet (set1 or set2)
   • Make sure to have intel realsense ROS package; if not, install sudo apt-get install ros-$ROS_DISTRO-realsense2-camera
   • Launch camera roslaunch realsense2_camera rs_camera.launch on different terminal

3. Keep the lights ON in the room

4. Use different surfaces

   • 

5. Use DECOY objects with our at-work objects, as shown in the image

   • 

6. Regarding where and how to place objects on the workstation

   • Where: keep objects in the 20 cm range on the workstation [check the below image]
   • How: keep at least 2 cm gap between the objects [check the below image]
   • 

[kvnptl]

• IDs 308 to 461, dataset collection from set 1 and set 2 completed and updated HERE.

[kvnptl]

• Use file_renaming.py script for renaming files

[kvnptl]

[UPDATE]

The initial YOLOv8(s) model trained on image ids 308-461 (both sets) is ready to use for further annotation. Please find the trained model HERE.

In order to use the custom trained model in anylabeling tool,

1. Download the ONNX and YAML files from HERE
   • NOTE: keep both files in the same directory
2. In anylabeling tool, click on "Auto Labeling" icon -> from the drop-down menu, select "Load Custom Model" -> select the downloaded YAML file.
3. Don't forget to verify the annotations and correct them (labels and bbox) if required

Please let me know if you face any issues.

[gokulkchenchani]

• 0 - 154 set 2 annotations completed and updated HERE.

[kvnptl]

[FINAL UPDATE]

• Dataset collection completed
• Dataset annotation completed

=> Find the full dataset HERE