search

TNTWEN / OpenVINO-YOLOV4

This is implementation of YOLOv4,YOLOv4-relu,YOLOv4-tiny,YOLOv4-tiny-3l,Scaled-YOLOv4 and INT8 Quantization in OpenVINO2021.3
MIT License
240 stars 66 forks source link

how to convert yolov4-tiny-3l model ? #20

Closed zhongwenkun886 closed 2 years ago

zhongwenkun886 commented 3 years ago

hi , I have trained the yolov4-tiny-3l model , but how to convert it to openVINO? thanks.

TNTWEN commented 3 years ago

Hi! Since tensorflow1.x based on static graph design is used, if the model is adjusted, tensorlow code needs to be rewritten. You need to modify these parts according to the model you are using: https://github.com/TNTWEN/OpenVINO-YOLOV4/blob/f7d066c56c5878dac2a9cc8603cb773e7c1dcd42/yolo_v4_tiny.py#L79-L109 https://github.com/TNTWEN/OpenVINO-YOLOV4/blob/f7d066c56c5878dac2a9cc8603cb773e7c1dcd42/yolo_v4_tiny.json#L10-L11

This means that you need to try to understand and adjust the tensorflow part of the code.For me, if we want to deploy YOLO on the embedded edge, I prefer to optimize the full version of the model. For the full yolov3/v4（not tiny）model, I made a script that automatically generates tensorflow code https://github.com/TNTWEN/OpenVINO-YOLO-Automatic-Generation. After November 9th, I will share how to optimize the full version of the model to get several times faster on OpenVINO without losing accuracy.

zhongwenkun886 commented 3 years ago

thinks

TNTWEN commented 3 years ago

@zhongwenkun886 https://github.com/TNTWEN/OpenVINO-YOLOV4/tree/v4-tiny-3l