TreB1eN

应该是pytorch的版本问题，你用的是老版本？

升级一下呢，我记得我遇到过这个问题，是版本的问题，你也可以google一下

@PCML-WangZhe I think 2 fps is quite fast indeed, what machine are you using btw

@PCML-WangZhe The backbone is Resnet101, How fast would you think it could be ? In the original paper, resnet101 is used to prove its' performance, another xception like backbone is...

@PCML-WangZhe what's the backbone, which repo, and test image size and proposals number in your comparison, this is not a speed oriented implementation, I dont think there is much point...

@jinfagang 100fps, according to the paper

@PCML-WangZhe this repo Test size is 1000

firstly, could you try not to use the precise mode? Because in precise mode the inference will be running multiple times, each time on a different input scale. Also, the...

really ? What I recall the step2 is pretty fast, I may need a while for me to test it again in my computer

模型加速是真的很难，不过有个[lightweight openpose](https://arxiv.org/pdf/1811.12004.pdf)