Manually adding new bounding boxes based on the position of darknet predictions. #8390
-
|
Hi there, I have used darknet in order to identify tomatoes as being Red, Green, Orange and Batch (Group of 8+ tomatoes). However, while yolo is successful in identifying the individual tomatoes the Batch is not very accurate. I think what would produce a more accurate model is identifying when there is a cluster of tomatoes together and then drawing a new bounding box from the top left most x,y coordinate of all the identified tomatoes to the bottom right most x,y coordinate. Any suggestions on where to start in order to achieve this. Can I get a tabular data of the YOLO predictions and then edit this? Would appreciate any help. Many thanks, Orlando |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
Search for "Pseudo-labelling" in the readme. Is that what you want? Your batch idea does work. I use it in my "city" project, where my classes include "many vehicles": https://www.youtube.com/watch?v=I-79ff1TD5M&t=24s The other thing that may be of interest to you is the JSON output if you are trying to identify where you can go back and label some "batch" tomatoes: https://www.ccoderun.ca/programming/darknet_faq/#json_output |
Beta Was this translation helpful? Give feedback.
-
|
Thank you. |
Beta Was this translation helpful? Give feedback.
Search for "Pseudo-labelling" in the readme. Is that what you want?
Your batch idea does work. I use it in my "city" project, where my classes include "many vehicles": https://www.youtube.com/watch?v=I-79ff1TD5M&t=24s
The other thing that may be of interest to you is the JSON output if you are trying to identify where you can go back and label some "batch" tomatoes: https://www.ccoderun.ca/programming/darknet_faq/#json_output