mindspore-lab / mindocr

A toolbox of OCR models, algorithms, and pipelines based on MindSpore
https://mindspore-lab.github.io/mindocr/
Apache License 2.0
174 stars 44 forks source link

Bugfix of svtr when the input is FP16 or FP32 for MindSpore r2.3rc1 #686

Closed Bourn3z closed 3 months ago

Bourn3z commented 3 months ago

Thank you for your contribution to the MindOCR repo. Before submitting this PR, please make sure:

Motivation

解决已知问题:当amp_level = 'O2'时,即网络输入为FP16时,grid_sample有计算问题。因此将算子输入手动转换到FP64计算。 临时规避方案,待r2.3.0修复grid_sample计算问题后,应回退该PR。 Solve known issues: When amp_level = 'O2', that is, when the network input is FP16, grid_sample has calculation problems. Therefore, the input is manually cast to FP64.

(Write your motivation for proposed changes here.)

Test Plan

(How should this PR be tested? Do you require special setup to run the test or repro the fixed bug?)

Related Issues and PRs

(Is this PR part of a group of changes? Link the other relevant PRs and Issues here. Use https://help.github.com/en/articles/closing-issues-using-keywords for help on GitHub syntax)