search

HRNet / HRNet-Semantic-Segmentation

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
Other
3.09k stars 682 forks source link

Why did I get better results than the paper #253

Open alexanderuo opened 2 years ago

alexanderuo commented 2 years ago

508E_` V`{29CY41BQ62~R Pre-training model ：hrnet_ocr_cs_trainval_8227_torch11.pth Maybe because of a previous error：The file could not be found ：final_state.pth I changed the name of the hrnet_ocr_cs_trainval_8227_torch11.pth model to this one and put it in the corresponding path. Meanwhile, I would like to ask how to get final_state.pth ,Maybe there's something wrong here

ydhongHIT commented 2 years ago

508E_ V{29CY41BQ62~R Pre-training model ：hrnet_ocr_cs_trainval_8227_torch11.pth Maybe because of a previous error：The file could not be found ：final_state.pth I changed the name of the hrnet_ocr_cs_trainval_8227_torch11.pth model to this one and put it in the corresponding path. Meanwhile, I would like to ask how to get final_state.pth ,Maybe there's something wrong here

Hi, I think that the trainval model was trained on both train and val set. Thus, the eval result on the val set is high. You should test it on the test set or evaluate the models trained on only train set.