Windows Support

[achanin00]

Can this project be built under Windows?

deploy_cpu.sh and deploy_trt.sh appear to be Linux only.

[SthPhoenix]

Hi! This project wasn't tested on Windows, but i think you can run CPU version without docker if you install all requirements from src/requirements.txt

CPU version uses onnxruntime for inference, but there is also a onnxruntime-gpu which should work on Windows, but you should also modify src/api_trt/modules/model_zoo/exec_backends/onnxrt_backend.py by replacing lines 12, 39, 80 with:
self.rec_model = onnxruntime.InferenceSession(rec_name, providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])

and line 117 with:
self.rec_model = onnxruntime.InferenceSession(model, providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])

Then use src/api_trt/entrypoint.sh as reference for commands required to start the service.

TRT version won't work on Windows because AFAIK there is no TRT Python wrapper for windows.

Also there were reports that at least CPU version works on Windows using Docker in WLS2, GPU version wasn't working last year, but maybe NVIDIA or Microsoft has fixed compatibility issues over the year

[achanin00]

Thank you for the pointers. I will give it a try.

…

On Fri, Feb 3, 2023, 3:45 PM SthPhoenix ***@***.***> wrote: Hi! This project wasn't tested on Windows, but i think you can run CPU version without docker if you install all requirements from src/requirements.txt CPU version uses onnxruntime for inference, but there is also a onnxruntime-gpu which should work on Windows, but you should also modify src/api_trt/modules/model_zoo/exec_backends/onnxrt_backend.py <https://github.com/SthPhoenix/InsightFace-REST/blob/master/src/api_trt/modules/model_zoo/exec_backends/onnxrt_backend.py> by replacing lines 12, 39, 80 with: self.rec_model = onnxruntime.InferenceSession(rec_name, providers=['CUDAExecutionProvider', 'CPUExecutionProvider']) and line 117 with: self.rec_model = onnxruntime.InferenceSession(model, providers=['CUDAExecutionProvider', 'CPUExecutionProvider']) Then use src/api_trt/entrypoint.sh <https://github.com/SthPhoenix/InsightFace-REST/blob/master/src/api_trt/entrypoint.sh> as reference for commands required to start the service. TRT version won't work on Windows because AFAIK there is no TRT Python wrapper for windows. Also there were reports that at least CPU version works on Windows using Docker in WLS2, GPU version wasn't working <#58> last year, but maybe NVIDIA or Microsoft has fixed compatibility issues over the year — Reply to this email directly, view it on GitHub <#110 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AYLQVBY4XXAD4P4T74JDZN3WVVU55ANCNFSM6AAAAAAUPQVZJI> . You are receiving this because you authored the thread.Message ID: ***@***.*** com>