search

scaelles / OSVOS-TensorFlow

One-Shot Video Object Segmentation
http://vision.ee.ethz.ch/~cvlsegmentation/osvos/
GNU General Public License v3.0
438 stars 132 forks source link

Parent model #6

Closed j50888 closed 7 years ago

j50888 commented 7 years ago

Hi, I wanna get the parent model(w/o boundary snapping) which reach 77.4 mioU I used the provided model(which finetuned 2k iterations) only reach 76.3 mioU Is there any training policy to reproduce the 77.4 mioU performance? Or could you provide the model? Thank you!

scaelles commented 7 years ago

Hello, The model that reaches 77.4 mIoU is the Caffe model provided here. The performance of the Tensorflow model is slightly lower. I have replicated the forward pass of the Caffe and Tensorflow models so I guess the difference in the final mIoU may be in the training phase.

j50888 commented 7 years ago

Thanks