# yoloe_data_engine

**Repository Path**: mirrors_ultralytics/yoloe_data_engine

## Basic Information

- **Project Name**: yoloe_data_engine
- **Description**: yoloe_data_engine
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2025-11-16
- **Last Updated**: 2026-02-08

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README




#  pipline of the data engine
###   read the grounding data from json file
    for each sample, per-store the others samples sharing the same image.
    add the 

###   model predict and save the jons files
-     visual to check the json files
-  found that some boxes are overlapped heavily, with different text
- how to deal with these boxes? 

###  merge model prediction to label,
- discard the bbox with higher iou  ( > 0.8, higher iou , no consider the class or text)





-    generate the visual prompt embedding for each instance (bbox)


-    merge bboxes within the same image ( consider the vpe distance  and text similarity ,bbox iou<0.8 )


-    transfer to grounding format cache for training

- 


to do:

write a tools to visual the bbox ious. within the same images