Pretrained models for YouTube-VOS dataset
XiaoqiangZhou opened this issue · comments
XiaoqiangZhou commented
Thanks for sharing such great work about video object segmentation and providing the detailed instructions for the code.
- Unfortunately, I find that the link for pretrained YouTube-VOS model is invalid. Have you finished retraining it? I'm looking forward to this model.
- A question about the training detail. For the training on the YouTube-VOS dataset, do you use the annotated frames or all the frames in the training set? As we know, the training set of this dataset is annotated every 5 frames, e.g. for the image id from 0 to 9, the No.0 and No.5 images are annotated with segmentation masks.
Thanks!
Haozhe Xie commented
@XiaoqiangZhou
Sorry, I am too busy to finish it.
I just use segmented frames (i.e. every 5 frames) for the YouTube-VOS dataset.