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

question about training my own data #23

Closed Adarine99 closed 6 years ago

Adarine99 commented 6 years ago

Hi, Appreciate your remarkable work and I want to train my own video data,but how can I get the first Labeled image,Is there any software I can use?Thanks a lot.

Adarine99 commented 6 years ago

got it ,I'm using the GIMP.

scaelles commented 6 years ago

Good to hear that you found a fix yourself! Otherwise, you could also use the LabelMe project.

Minotaur-CN commented 5 years ago
  1. The label is 3-channel png? or has a palette as VOC?
  2. Every sampling video has to train a model? Thanks!
kmaninis commented 5 years ago
  1. It's a 1-channel binary image.
  2. That's correct. Every video needs online training.
Minotaur-CN commented 5 years ago

Thanks for your reply! I tested some video not from DAVIS, The evaluation results not well, even use on-line training for a long training steps. How to improve the results? Re-train the parent work?

Thanks!