Closed qiulesun closed 1 year ago
mIoU
@DerrickWang005 As far as I know, current SOTAs, such as [1,2,3], mainly report oIoU value for comparisons. So do you consider providing corresponding oIoU values across all benchmark datasets?
[1] LAVT: Language-Aware Vision Transformer for Referring Image Segmentation, In CVPR 2022 [2] ReSTR: Convolution-free Referring Image Segmentation Using Transformers, In CVPR 2022 [3] Bottom-Up Shift and Reasoning for Referring Image Segmentation, In CVPR 2021
Nice work ! In terms of IoU reported in CRIS, the IoU is overall intersection-over-union (oIoU) or mean intersection-over-union (mIoU) metric? In the released code, the mIoU metric seems to be calculated for comparisons with SOTA in all tables in CRIS paper.