Wenfei Zhu
Wenfei Zhu
@liqikai9 I am currently using this modification `batch_size = len(data['img_metas'].data[0])` for my own test, but I am not sure if this is general enough.
Could you please start developing this function? I think this function is very useful especially when auto-annotation produces a lot of useless annotations that you want to delete.
我也对比了一下cpu和gpu,发现几个点: 1. 每张图第一次跑都很慢,第二次就快了。而且这和pytorch那种第一次跑比较慢不一样,这是每张图的第一次都慢。举个栗子,按下列顺序测试图片: - 图片1,需要跑几秒钟 - 图片1,只要跑0.几秒 - 图片2,需要跑几秒钟 - 图片2,只要跑0.几秒 - 图片1,只要跑0.几秒 2. 如果不看一张图的第一次运行时间,而只看第二次的话,可以发现: - 目标检测gpu比cpu快 - 角度检测和文字识别gpu比cpu慢 感觉这里面有不少内存和显存之间数据搬运的问题。可能作者写的时候就没有给gpu优化?
Same problem here. The configuration below does not work. ```bash ipympl 0.2.1 py36_1001 conda-forge ipython 7.10.2 py36h39e3cac_0 ipywidgets 7.5.1 py_0 conda-forge jupyterlab 1.0.2 py36hf63ae98_0 matplotlib 3.1.1 py36h5429711_0 nodejs 10.13.0 he6710b0_0...
> I fixed this with [#130 (comment)](https://github.com/matplotlib/jupyter-matplotlib/issues/130#issuecomment-567931502) your solution truly works, thank you!
> I am not 100% sure, but I imagine that you need to update https://github.com/NVlabs/few_shot_gaze/blob/master/demo/monitor.py#L28 (and its inverse) based on these values that you determined: > > ``` > T...
> > > I am not 100% sure, but I imagine that you need to update https://github.com/NVlabs/few_shot_gaze/blob/master/demo/monitor.py#L28 (and its inverse) based on these values that you determined: > > >...
这个不是huggingface。web_demo默认用的是modelscope,我也遇到了楼主的问题。Download interval is too small 似乎是modelscope的一个问题,在他们的issue里能找到,目前还没解决,但是似乎降低modelscope的版本能解决这个问题。 另外就是,挂个代理,换成huggingface,也能解决这个问题。 > 你这是在huggingface上下载模型报错了,因为huggingface被墙了,建议自己开梯子手动把模型下载下来,然后加载local path
> > 这个不是huggingface。web_demo默认用的是modelscope,我也遇到了楼主的问题。Download interval is too small 似乎是modelscope的一个问题,在他们的issue里能找到,目前还没解决,但是似乎降低modelscope的版本能解决这个问题。 > > 另外就是,挂个代理,换成huggingface,也能解决这个问题。 > > > 你这是在huggingface上下载模型报错了,因为huggingface被墙了,建议自己开梯子手动把模型下载下来,然后加载local path > > 换成hf之后,你成功infer了嘛 可以infer的
I also need an easier way to change the layer of an object, because I want to label the relationship such as A is occluded by B and B is...