zhanggang001 / RefineMask

RefineMask: Towards High-Quality Instance Segmentation with Fine-Grained Features (CVPR 2021)

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

mmdetection vs. detectron setting

SangbumChoi opened this issue · comments

Hi thanks for the nice work.
I just wonder after seeing this issue SysCV/transfiner#11 (comment) it makes me to ask two questions

  1. Is there any model from Table 8 (SOTA comparison) trained on detectron2?
  2. Also normally mmdetection is run by EpochBasedRunner and detectron2 is run by iteration. (Additionally many different hyperparameter setting) Do you think default setting comapring between two libraray is fair comparison?

Hi thanks for the nice work. I just wonder after seeing this issue SysCV/transfiner#11 (comment) it makes me to ask two questions

  1. Is there any model from Table 8 (SOTA comparison) trained on detectron2?
  2. Also normally mmdetection is run by EpochBasedRunner and detectron2 is run by iteration. (Additionally many different hyperparameter setting) Do you think default setting comapring between two libraray is fair comparison?
  1. The models with ‡ were trained with a similar setting to detectron2 using mmdetection. The last sentence of the caption of Table 8 states this.
  2. Most of the models in Table 1 and Table 8 we compared with (including Mask R-CNN, PointRend, HTC) were trained by ourselves with the same setting using mmdetection.