search

SUYEgit / Surgery-Robot-Detection-Segmentation

Object detection and segmentation for a surgery robot using Mask-RCNN on Python 3, Keras, and TensorFlow..
Other
181 stars 83 forks source link

How you modify the code to reach the real-time performance? #10

Open fengsky401 opened 6 years ago

fengsky401 commented 6 years ago

I use mask-rcnn(backbone resnet-101) to train and predict my image, when I predict the image, only model.detect spends 250ms-300ms. I want to reduce the time in model.detect to 200-250ms, I notice you also use resnet-101 as backbone, and wonder how you modify the details in the code. Thank you very much!

759420184 commented 5 years ago

I use mask-rcnn(backbone resnet-101) to train and predict my image, when I predict the image, only model.detect spends 250ms-300ms. I want to reduce the time in model.detect to 200-250ms, I notice you also use resnet-101 as backbone, and wonder how you modify the details in the code. Thank you very much!

Hello, my machine's GPU is 1080Ti, I was trained my own dataset and run it use ~750ms per frame. Is the time of detection used related to the GPU? How is your machine detection speed so fast? Looking forward to your reply, thank you!

pengyuchu commented 5 years ago

I use mask-rcnn(backbone resnet-101) to train and predict my image, when I predict the image, only model.detect spends 250ms-300ms. I want to reduce the time in model.detect to 200-250ms, I notice you also use resnet-101 as backbone, and wonder how you modify the details in the code. Thank you very much!

Hello, my machine's GPU is 1080Ti, I was trained my own dataset and run it use ~750ms per frame. Is the time of detection used related to the GPU? How is your machine detection speed so fast? Looking forward to your reply, thank you!

Sometimes it's also related to your input resolution. Even if you use same GPU to do the inference work, distinct resolution will cause various cost. Try to reduce the resolution of your input and perhaps it can work on your case.