The OCR approach is rephrased as Segmentation Transformer: https://arxiv.org/abs/1909.11065. This is an official implementation of semantic segmentation for HRNet. https://arxiv.org/abs/1908.07919
I found there is code for calculating GFLOPs within the repository. Is this the code that was used to calculate HRNet GFLOPs written in Table 3 of the paper? Because judging by the code, it converts FLOPs to GFLOPs by dividing 1024 ^ 3. But as I know you have to divide it by 1000 ^ 3.