FastSAM tensorrt inference?

Jun 26 '23 05:06 machine52vision

I converted pt to onnx, and to trt. But outputs infer are 2 tensors have shape (1,32,160,160) , (1,32,8400). I couldn't post process to get mask. Please help me.

Jun 26 '23 08:06 ChuRuaNh0

I handled it. You can reference at link: [GitHub] (https://github.com/ChuRuaNh0/FastSam_Awsome_TensorRT). If it helpful with you, give me a star, thanks.

Jun 27 '23 04:06 ChuRuaNh0

@ChuRuaNh0 thank you, you saved ton of time

Jun 27 '23 05:06 thanhson28

@ChuRuaNh0 thank you! For better performance, add an "INMSLayer" at the end of network is a better choice!

Jun 28 '23 07:06 yuanjiechen

I handled it. You can reference at link: [GitHub] (https://github.com/ChuRuaNh0/FastSam_Awsome_TensorRT). If it helpful with you, give me a star, thanks.

Thanks for your contribution! We will add FastSAM_Awesome_TensorRT link to README.md, so that more people can benefit from your contribution.

Jun 29 '23 01:06 an-yongqi

@AAAAAAAyq I'm appreciate. It's great to hear that.

Jun 29 '23 12:06 ChuRuaNh0

I handled it. You can reference at link: [GitHub] (https://github.com/ChuRuaNh0/FastSam_Awsome_TensorRT). If it helpful with you, give me a star, thanks.

Great Work! I want to know the inference speed and your device when you use the ONNX or TRT. Looking forward to your reply

Jul 01 '23 09:07 FT115