search

carrierlxk / AGNN

Zero-shot Video Object Segmentation via Attentive Graph Neural Networks (ICCV2019 Oral)
207 stars 25 forks source link

About code for Multi Object Inference #10

Open IIAT-MR-LL opened 4 years ago

IIAT-MR-LL commented 4 years ago

@carrierlxk Hi, Xiankai, I am confused about the inference procedure on DAVIS17 and YouTubeVOS, would you mind sharing the inference or test code for multi object segmentation?

Thanks.

carrierlxk commented 4 years ago

Thanks for your interesting in your work. The multi object segmentation codes comes from PReMVOS. We use bi-directional optical flow and the AGNN output to identify the instances in the first frame, then perform video object segmentation in the subsequent frame as PReMVOS.