Closed warmspringwinds closed 7 years ago
Hi, I did it long time ago relying on this FAQ by the DeepLab team: http://ccvl.stat.ucla.edu/deeplab_faq/ Now it seems that the link provided to the dataset with extra annotations is broken. I will try to find an alternative link.
https://github.com/martinkersner/train-DeepLab
This GitHub branch provide the scripts on how to build the augmented dataset
By the way, 76.5% is the result from the paper or your tensorflow code?
76.5% is the result of the converted to TF model (the link to the model file is provided in README).
Thanks. Note that the segmentation labels are currently distributed as .mat files. PR #4 adds a simple script to generate images that can be used with the current trainer.
Can you tell me the mAP if the model is trained from scratch?
Unfortunately, I cannot tell you that. PRs are always welcome.
Hi.
Thank you for your repository!
I have a small question: where did you get the augmented dataset from?