Hello,What do these parameters (fpn_levels_0/1, gt_inds_0/1, inst_im_inds,inst_rel_ids) mean in crossvis.py

hustvl / CrossVIS

[ICCV 2021] Crossover Learning for Fast Online Video Instance Segmentation

https://arxiv.org/abs/2104.05970

Other

85 stars 4 forks source link

Hello,What do these parameters (fpn_levels_0/1, gt_inds_0/1, inst_im_inds,inst_rel_ids) mean in crossvis.py #15

Closed Alxx999 closed 2 years ago

vealocia commented 2 years ago

Hi, @Alxx999! Thanks for your attention in our work. fpn_levels and gt_inds indicates the corresponding fpn feature and ground truth for each predicted instance, respectively. You can try running CondInst to get details. As we sample a pair of frames from one video in a batch, we use 0 and 1 to identify two frames. inst_im_inds and inst_rel_ids are used to find the instance relations between a pair of frames.

Alxx999 commented 2 years ago

你好，@Alxx999！感谢您对我们工作的关注。 fpn_levels并gt_inds分别表示每个预测实例对应的 fpn 特征和ground truth。您可以尝试运行CondInst以获取详细信息。当我们从一个视频中批量采样一对帧时，我们使用 0 和 1 来识别两个帧。 inst_im_inds并inst_rel_ids用于查找一对帧之间的实例关系。

OK！Thank you very much!