Guillaume Infantes

detection : cannot handle filenames with a " "

1

when parsing train.txt , inputconnectors parses per line stuff like "img.jpg bbox.txt" , it cannot handle "img dummy.png bbox dummy.png" see https://github.com/jolibrain/deepdetect/blob/master/src/backends/caffe/caffeinputconns.cc#L788 at very least it should `continue` instead of...

optional support for cuda enabled opencv

1

input connector 's resize is done only on CPU, and takes more time then inference on many GPU enabled platforms. When available, cv::resize should be replaced by cv::cuda::resize of cv::gpu::resize...

- googlenet : caffe2ncnn is not able to parse googlenet.caffemodel (???!!) - se_resnet_50 : axpy layer is not implemented in our version of ncnn : a fix for external caffe2ncnn...

here is a list: - shfl_down ``` src/caffe/3rdparty/ctc/reduce.cu(44): warning: function "__shfl_down(float, unsigned int, int)" /usr/local/cuda/include/sm_30_intrinsics.hpp(275): here was declared deprecated ("__shfl_down() is deprecated in favor of __shfl_down_sync() and may be removed...

problem in batching wrt schemes and subgraphs

3

## 🐛 Bug When batching graphs, schemes can be incorrectly checked, in the case were the scheme is present but the number of edge with this scheme is 0 ##...