onnx to tflite code conversion and inferencing

[Frontkick]

Hello Goutam,

first of all your work on object tracking is amazing

I am currently trying to inference original onnx runtime code to tflite but having some issues:
I loaded tflite model in the tensorflow interpreter and did preprocessing of image with model similar to onnx but having some issues in output shape of output variables which are given below

Variable name (Shape in onnx) (Shape in Tflite)
out_score_map (1,1,16,16) (1,16,16,2)
out_size_map (1,2,16,16) (1,16,16,2)
out_offset_map (1,2,16,16) (1,1,4)

is different than that of onnx model

specially of out_size_map, shape of out_size_map in onnx model is (1,2,16,16) but in tflite it is coming as (1,1,4)

here is the link of Repo from where i converted onnx model to tflite: https://github.com/MPolaris/onnx2tflite
here is the link of Repo which i'm working on: https://github.com/Frontkick/MVT_Tflite

you can just clone repository and run run.py file and can see where issue is

[Frontkick]

Hi Goutam,

I've successfully performed inference using the TFLite model and obtained output, but the results are different from those of the ONNX model. Here's the link to the TFLite model inference repository: https://github.com/Soumylenviz/mvt_tflite

you can run the code using run.py file

Bounding Box which im getting as ouput is totally different and is covering whole video, not tracking the actual object

Could you help me resolve the issue?