From 894e02df6f67e0b12ad1259abecd4b0543c55ac2 Mon Sep 17 00:00:00 2001 From: pytorchsam <72172677+pytorchsam@users.noreply.github.com> Date: Thu, 27 May 2021 17:37:58 -0500 Subject: [PATCH] Update 2021-5-26-torchvision-mobilenet-v3-implementation.md --- _posts/2021-5-26-torchvision-mobilenet-v3-implementation.md | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/_posts/2021-5-26-torchvision-mobilenet-v3-implementation.md b/_posts/2021-5-26-torchvision-mobilenet-v3-implementation.md index 5d11f70bda6a..6496c42a1806 100644 --- a/_posts/2021-5-26-torchvision-mobilenet-v3-implementation.md +++ b/_posts/2021-5-26-torchvision-mobilenet-v3-implementation.md @@ -120,7 +120,8 @@ In this section, we will first provide benchmarks of the released models, and th Here is how the models are initialized: ``` -high_res = torchvision.models.detection.fasterrcnn_mobilenet_v3_large_fpn(pretrained=T low_res = torchvision.models.detection.fasterrcnn_mobilenet_v3_large_320_fpn(pretraine +high_res = torchvision.models.detection.fasterrcnn_mobilenet_v3_large_fpn(pretrained=True) +low_res = torchvision.models.detection.fasterrcnn_mobilenet_v3_large_320_fpn(pretrained=True) ``` Below are some benchmarks between new and selected previous models. As we can see the high resolution Faster R-CNN with MobileNetV3-Large FPN backbone seems a viable replacement of the equivalent ResNet50 model for those users who are willing to sacrifice few accuracy points for a 5x speed-up: @@ -166,7 +167,8 @@ In this section we will start by providing some benchmarks of the released pre-t This is how to initialize the pre-trained models: ``` -lraspp = torchvision.models.segmentation.lraspp_mobilenet_v3_large(pretrained=True) deeplabv3 = torchvision.models.segmentation.deeplabv3_mobilenet_v3_large(pretrained=Tr +lraspp = torchvision.models.segmentation.lraspp_mobilenet_v3_large(pretrained=True) +deeplabv3 = torchvision.models.segmentation.deeplabv3_mobilenet_v3_large(pretrained=True) ``` Below are the detailed benchmarks between new and selected existing models. As we can see, the DeepLabV3 with a MobileNetV3-Large backbone is a viable replacement of FCN with ResNet50 for the majority of applications as it achieves similar accuracy with a 8.5x speed-up. We also observe that the LR-ASPP network supersedes the equivalent FCN in all metrics: