Segmentation using Pascal VOC Dataset is not giving result as per expected

dusty-nv / jetson-inference

Hello AI World guide to deploying deep-learning inference networks and deep vision primitives with TensorRT and NVIDIA Jetson.

https://developer.nvidia.com/embedded/twodaystoademo

MIT License

7.59k stars 2.94k forks source link

Segmentation using Pascal VOC Dataset is not giving result as per expected #162

Closed nicks165 closed 1 year ago

nicks165 commented 6 years ago

Hi I have used FCN-ALexnet model on PASCAL VOC Dataset. I got the pretrained model on Berkeley Vision. outputblob is sscore_fr_21 and input is data. However it seems the argmax calculated on Overlay is 0. And also the image just has another color on top of it. It did not do any segmentation. It is same problem both on JEtson TX2 and Desktop (1060)

S4WRXTTCS commented 6 years ago

My understanding is it's a known issue.

It's been mentioned a handful of times, but there isn't any fix

https://devtalk.nvidia.com/default/topic/1021784/jetson-tx2/unable-to-get-segmentation-to-work-with-jetson-tx2/post/5201703/#5201703 https://github.com/dusty-nv/jetson-inference/issues/87 https://github.com/dusty-nv/jetson-inference/issues/116