feature extract error

[budaLi]

[wenmengzhou]

can you attach the exported checkpoint and give the model info

[budaLi]

看起来是训练时候的配置参数是OrderDict类型的，不支持直接写入

[budaLi]

can you attach the exported checkpoint and give the model info

请问有比较好的解决方法嘛

[wenmengzhou]

can you attach the exported checkpoint and give the model info

请问有比较好的解决方法嘛

看起来你是直接用的训练保存的checkpoint做预测，而没有进行export模型？

[MuhammadHakami]

exported the model and still faced the same issue. it was due to the TorchFeatureExtractor writing the dictionary model configs into a disk while expecting it as a string. for a quick walkaround, just replace "ofile.write(config_str)" in line 48 with:

import json
json.dump(config_str, ofile)

silly issue. maybe some other models have the value config checkpoint['meta']['config'] as a string.