search

WongKinYiu / yolov9

Implementation of paper - YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information
GNU General Public License v3.0
8.54k stars 1.3k forks source link

Will there be any "T" class model for Instance Segmentation? #514

Open fukingbus opened 1 week ago

fukingbus commented 1 week ago

Greetings, Im working on mobile platform with Tensorflow Lite and the gelan-c segmentation model is doing well, but with 2000-3000ms inference time. Will there be any "T" class training model and pre-trained weight releasing soon? or any help to self modifying a smaller parameter model designed for mobile purpose? :(

dsbyprateekg commented 1 week ago

@fukingbus how have you converted the gelan-c segmentation model to the tflite format?

fukingbus commented 1 week ago

@fukingbus how have you converted the gelan-c segmentation model to the tflite format?

Apparently its not possible to do direct export in tflite format right now , either your model was trained using gelan-c or yolov9 dual branch. #250 #472 But it turns out you actually can export gelan-c trained model in onnx format using export.py

after exporting to onnx format you are able to do onnx to tensorflow conversion https://github.com/PINTO0309/onnx2tf

and after tensorflow conversion its easy to convert to tensorflow-lite

dsbyprateekg commented 1 week ago

Thanks @fukingbus . It seems the repo https://github.com/PINTO0309/onnx2tf is the silver bullet for tf conversion. Author has done the great work.