diff --git a/docker/tensorrt/requirements-amd64.txt b/docker/tensorrt/requirements-amd64.txt index dd99190d0..b5ad4fcbd 100644 --- a/docker/tensorrt/requirements-amd64.txt +++ b/docker/tensorrt/requirements-amd64.txt @@ -7,7 +7,7 @@ cython == 0.29.*; platform_machine == 'x86_64' nvidia-cuda-runtime-cu12 == 12.1.*; platform_machine == 'x86_64' nvidia-cuda-runtime-cu11 == 11.8.*; platform_machine == 'x86_64' nvidia-cublas-cu11 == 11.11.3.6; platform_machine == 'x86_64' -nvidia-cudnn-cu11 == 8.5.0.*; platform_machine == 'x86_64' +nvidia-cudnn-cu11 == 8.6.0.*; platform_machine == 'x86_64' nvidia-cufft-cu11==10.*; platform_machine == 'x86_64' onnx==1.14.0; platform_machine == 'x86_64' onnxruntime-gpu==1.17.*; platform_machine == 'x86_64' diff --git a/frigate/detectors/plugins/onnx.py b/frigate/detectors/plugins/onnx.py index 1939b7323..bf5c0be68 100644 --- a/frigate/detectors/plugins/onnx.py +++ b/frigate/detectors/plugins/onnx.py @@ -36,7 +36,28 @@ class ONNXDetector(DetectionApi): path = detector_config.model.path logger.info(f"ONNX: loading {detector_config.model.path}") - self.model = ort.InferenceSession(path, providers=ort.get_available_providers()) + + providers = ort.get_available_providers() + options = [] + + for provider in providers: + if provider == "TensorrtExecutionProvider": + options.append( + { + "trt_timing_cache_enable": True, + "trt_timing_cache_path": "/config/model_cache/onnx", + "trt_engine_cache_enable": True, + "trt_engine_cache_path": "/config/model_cache/onnx/trt-engines", + } + ) + elif provider == "OpenVINOExecutionProvider": + options.append({"cache_dir": "/config/model_cache/onnx"}) + else: + options.append({}) + + self.model = ort.InferenceSession( + path, providers=providers, provider_options=options + ) self.h = detector_config.model.height self.w = detector_config.model.width