pptsm 模型推理时报错

PaddlePaddle / PaddleVideo

Awesome video understanding toolkits based on PaddlePaddle. It supports video data annotation tools, lightweight RGB and skeleton based action recognition model, practical applications for video tagging and sport action detection.

Apache License 2.0

1.54k stars 383 forks source link

pptsm 模型推理时报错 #384

Open zidingyi-wangda opened 2 years ago

zidingyi-wangda commented 2 years ago

加载从 https://bj.bcebos.com/acg-algo/PaddleVideo_application/VideoQualityAssessment/ppTSM.pdparams 下载的预训练模型推理时，报 OSError: {weight_path} is not a checkpoint file 错误，该如何解决？

huangjun12 commented 2 years ago

看报错可能是路径设置不对

可以参考文档check下哈： https://github.com/PaddlePaddle/PaddleVideo/blob/develop/docs/zh-CN/model_zoo/recognition/pp-tsm.md#模型推理

geekchen007 commented 2 years ago

https://bj.bcebos.com/acg-algo/PaddleVideo_application/VideoQualityAssessment/ppTSM.pdparams ValueError: paddle.load can not parse the file:pretrained_dygraph/ppTSM.pdparams. 已经排除路径错误。 Debug信息：

预训练模型是140M，自己从头train的model是89.9M。环境：
paddlepaddle-gpu 2.2.2 py37_gpu_cuda10.1_many_linux

geekchen007 commented 2 years ago

参考文档的预训练模型可用 [https://github.com/PaddlePaddle/PaddleVideo/blob/develop/docs/zh-CN/model_zoo/recognition/pp-tsm.md#模型推理]