feature extract error
budaLi opened this issue · 5 comments
budaLi commented
wenmengzhou commented
can you attach the exported checkpoint and give the model info
budaLi commented
budaLi commented
can you attach the exported checkpoint and give the model info
请问有比较好的解决方法嘛
wenmengzhou commented
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 commented
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.