{"id":5558,"date":"2017-09-19T12:22:57","date_gmt":"2017-09-19T16:22:57","guid":{"rendered":"http:\/\/local.brightwhiz\/?p=5558"},"modified":"2017-09-19T12:22:57","modified_gmt":"2017-09-19T16:22:57","slug":"dlib-19-7-release-download","status":"publish","type":"post","link":"http:\/\/local.brightwhiz\/dlib-19-7-release-download\/","title":{"rendered":"Dlib 19.7 Release is now Available for Download"},"content":{"rendered":"
The new Dlib 19.7 Release has been announced and is now available for download. This new version comes with several improvements, new features and bug fixes over previous releases.<\/p>\n
Dlib<\/a> is a free and open source<\/a> machine learning toolkit. It is written in modern C++<\/a> an contains a collection of machine learning<\/a> algorithms. It is ideal for use in developing simple to complex C++ based software to solve real world problems in domains such as robotics, mobile computing, embedded systems<\/a>, high performance computing environments and more.<\/p>\n There are bug fixes centered around the Also fixed is a bug where Finally also silently fixed is the loss_per_missed_target parameter of the New features in this version that are centered around Deep learning. CNN+MMOD detectors have now doubled in speed as well as it now being a multi-class detector.<\/p>\n There are now improvements in the imglab tool that center around usability. There is also a new five point landmarking model which is faster and works with HOG and CNN. This makes it the preferred landmarking model to use.<\/p>\n Raw pointers can now be used in sub_images of anything. What’s Included in the Dlib 19.7 Release<\/h2>\n
input_rgb_image_pyramid::image_contained_point()<\/code> function. This bug caused the function to erroneously indicate that a point wasn’t inside the original image when really it was. This caused error messages.<\/p>\n
mmod_options<\/code> would pick bad window sizes in some cases. There is also a fix where the extract layer that triggered when a tensor with a different number of samples than the tensor used to initialize the network was passed through the layer.<\/p>\n
loss_mmod_<\/code> which was not being used right when boxes were auto-ignored .<\/p>\n
std::chrono::high_resolution_clock<\/code> is now implemented and therefore it supports much higher precision. Chinese whispers clustering has now been exposed to Python.<\/p>\n