mean_iu is not good enough

shekkizh / FCN.tensorflow

Tensorflow implementation of Fully Convolutional Networks for Semantic Segmentation (http://fcn.berkeleyvision.org)

MIT License

1.25k stars 527 forks source link

mean_iu is not good enough #98

Open worstprogrammerCN opened 6 years ago

worstprogrammerCN commented 6 years ago

I trained the model and calculated the miou(mean intersection over union)myself, but the miou is just around 19. According to the baseline in this page http://sceneparsing.csail.mit.edu, the baseline of miou of Fcn8s is 24. I have to do sth. to improve the miou. I have set the learning rate to 1e-6. I want to know if the author has tested the miou？What can I do to improve my miou，still using FCN-8s？ I'll be grateful to anyone who helps me！

greenhornno1 commented 5 years ago

hi , I want to test my model mean_iu as well I am totally new for tensorflow so i dont know how to edit the code , can you share yours ? Thank you