Model if fully Trained or Can you provide mobile Application Link

[Frontkick]

Hello ,
can you confirm that the model (Updated_Lighttrack.onnx) which you have used in your Application is fully trained and also
how much FPS does it give while tracking

Also can you please provide Mobile Application link of the model

Thank You

[Z-Xiong]

The model was exported from the official PyTorch model, and I did not develop an Android application. I only compiled and verified the program on the computer.
Thank you for your attention.

[Oleg-lcb]

The model was exported from the official PyTorch model, and I did not develop an Android application. I only compiled and verified the program on the computer. Thank you for your attention.

Thanks. But why did you change the size of input data from 256x256 to 288x288? I don't understand this thing. Very interested in this)

[Z-Xiong]

The model was exported from the official PyTorch model, and I did not develop an Android application. I only compiled and verified the program on the computer. Thank you for your attention.

Thanks. But why did you change the size of input data from 256x256 to 288x288? I don't understand this thing. Very interested in this)

There's no specific reason other than to adapt to the size of the original model after downsampling.

[Oleg-lcb]

The model was exported from the official PyTorch model, and I did not develop an Android application. I only compiled and verified the program on the computer. Thank you for your attention.

Thanks. But why did you change the size of input data from 256x256 to 288x288? I don't understand this thing. Very interested in this)

There's no specific reason other than to adapt to the size of the original model after downsampling.

I'm asking about size of tensor that you give to model's input backbone from testing image. In original model (from user researchmm) input is 256x256 tensor. But you use (change) input 288x288. May be I didn't understand your answer?

[Z-Xiong]

The model was exported from the official PyTorch model, and I did not develop an Android application. I only compiled and verified the program on the computer. Thank you for your attention.

Thanks. But why did you change the size of input data from 256x256 to 288x288? I don't understand this thing. Very interested in this)

There's no specific reason other than to adapt to the size of the original model after downsampling.

I'm asking about size of tensor that you give to model's input backbone from testing image. In original model (from user researchmm) input is 256x256 tensor. But you use (change) input 288x288. May be I didn't understand your answer?

https://github.com/researchmm/LightTrack/blob/39c426f48ee674795cdf0e00301a0b4ad0785d2a/experiments/test/VOT/LightTrack.yaml#L7

[Oleg-lcb]

Ok, I see, thanks!