Guang YANG
Guang YANG
demo.py里有行bug img = cv2.resize(img, (0, 0), fx=config.MODEL.IMAGE_SIZE.H / h, fy=config.MODEL.IMAGE_SIZE.H / h, interpolation=cv2.INTER_CUBIC) fx写错了
我按照readme直接把那个pooling去掉换成reshape： layer { name: "vertical_reshape" type: "Reshape" bottom: "BatchNorm3" top: "vertical_output" reshape_param { shape { dim: 0 dim: -1 dim: 1 dim: 0 } } } 然后训练无法收敛，是什么原因呀
update git works for me yum install git
The position of [EOT] token is different for text with diff length, this dose not confuse the learning of position embedding?
> Building engine, please wait for a while... > [04/16/2021-10:54:24] [E] [TRT] ../builder/cudnnBuilderWeightConverters.cpp (555) - Misc Error in operator(): 1 (Weights are outside of fp16 range. A possible fix is...
I use flownet2 to replace pyflow
> 如果想转换成trt支持的fp16或者Int8类型，应该如何修改。 目前默认是转成fp16的，见crnn_trt/crnn_number.cpp 里的 #define USE_FP16 （如果不想转，把这一行去掉就行），目前转INT8没有支持，等我弄一下
> 好的了解了，感谢你分享的代码，对我很有帮助。 感谢关注，等我把int8弄好了告诉你。
> Does it also work for traditional chinese? the network of course yes, but the pretrained weights is not supported for the traditional chinese, you need to train with the...
> @ygfrancois 输入图片32*100，直接使用pytorch推理时间为14.46ms，使用fp32量化是6.69ms，使用fp16量化是6.32ms。为什么fp32和fp16量化的差异这么小，这是否正常。 你用的显卡是2080ti吗？显存减少明显吗？我估计和硬件或者cudnn的内部实现有关