lianawang / itn Goto Github PK

@zlmzju Could you tell me the version of CUDA & CUDNN, I met an error about "pycuda", so I wanna know whether it's the version issue.

作者你好，我看文章中的网络结构是三个feature map 送入fusion层，而在Itn-vgg中只是把conv5_3送入fusion中，这两者有区别么？还是我没有理解清楚，忘指教。谢谢。

I use the ic15_itn_resnet.caffemodel to test the picture in ICDAR2015,the result is in outputfile,one is the foramt like 000001.txt 000002.txt and also generate a file called submit_roi ,so ,what the difference between them,i use the submit_roi file to draw the box in the origin picture,i got the following answer:

so what's wrong ,thanks

I run the test.sh in experiments/itn_vgg/, and using the weight ic15_itn_vgg.caffemodel, proto file experiments/itn_vgg/proto/test.pt

Cannot copy param 0 weights from layer 'fusion'; shape mismatch. Source param shape is 1024 1024 3 3 (9437184); target param shape is 512 512 3 3 (2359296). To learn this layer's parameters from scratch rather than copying from a saved net, rename the layer.

Does ic15_itn_resnet.caffemodel means train in the icdar2015 data and msra_itn_resnet.caffemodel means train in the MSRA data ? why not use icdar2015 and msra to trian a single model?
@zlmzju