tag:blogger.com,1999:blog-6061887630060661987.post754453959065098017..comments2024-03-13T13:54:35.259-04:00Comments on dlib C++ Library: High Quality Face Recognition with Deep Metric LearningDavis Kinghttp://www.blogger.com/profile/16577392965630448489noreply@blogger.comBlogger466125tag:blogger.com,1999:blog-6061887630060661987.post-13795406542839657032024-02-22T07:12:38.832-05:002024-02-22T07:12:38.832-05:00I am not making any claim of rights to the work I ...I am not making any claim of rights to the work I did here. As for what others might claim about datasets, I have no idea. You should talk to a lawyer if you have concerns.Davis Kinghttps://www.blogger.com/profile/16577392965630448489noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-90751050130445746612024-02-09T04:46:05.159-05:002024-02-09T04:46:05.159-05:00I am concerned about the sentence "the pretra...I am concerned about the sentence "the pretrained model used by this example program is in the public domain. So you can use it for anything you want."<br /><br />You used VGGFace and Face Scrub to train the model. And at least face scrub is definitely under a non-commercial license. So how can the resulting model be "used for everything you want"?Linko994https://www.blogger.com/profile/11620907999182732709noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-35300988986848681152023-06-22T21:48:57.076-04:002023-06-22T21:48:57.076-04:00Yep, that example program has the details.Yep, that example program has the details.Davis Kinghttps://www.blogger.com/profile/16577392965630448489noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-48597575783054801732023-06-20T20:40:29.317-04:002023-06-20T20:40:29.317-04:00i think the details of the 29 conv layer are in th...i think the details of the 29 conv layer are in this file <a href="https://devdoc.net/c/dlib-19.7/dnn_metric_learning_on_images_ex.cpp.html" rel="nofollow">dnn_metric_learning_on_images_ex.cpp</a> right?Achmad Rifkihttps://www.blogger.com/profile/07460986176914904282noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-50052404335742531522023-06-20T20:37:30.810-04:002023-06-20T20:37:30.810-04:00This comment has been removed by the author.Achmad Rifkihttps://www.blogger.com/profile/07460986176914904282noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-75449330661069234322023-06-20T11:15:24.013-04:002023-06-20T11:15:24.013-04:00Hello Davis, thanks for your works, such a great l...Hello Davis, thanks for your works, such a great library.<br /><br />so, i came into comment of your model details <b>For those interested in the model details, this model is a ResNet network with 29 conv layers. It's essentially a version of the ResNet-34 network from the paper Deep Residual Learning for Image Recognition by He, Zhang, Ren, and Sun with a few layers removed and the number of filters per layer reduced by half.</b><br /><br />i want to ask you, which layer did you remove from 34 conv layers to 29 conv layers, would you mind showing us the details of which layers removed? Thank youAchmad Rifkihttps://www.blogger.com/profile/07460986176914904282noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-17479492310244993822022-05-12T10:00:20.108-04:002022-05-12T10:00:20.108-04:00@Andyrey Thank you for your Answer!@Andyrey Thank you for your Answer!Anonymoushttps://www.blogger.com/profile/04493212271012455282noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-54748124409085947502022-05-12T09:40:56.101-04:002022-05-12T09:40:56.101-04:00@Unknown: zest of this method is that you don'...@Unknown: zest of this method is that you don't need to train the model for new faces. It is trained only to compare two faces and give the answer- is it the same person or not. But some ethnic bias is possible.<br /> Pre-treatment is necessary- the face should be large enough and not be smoothed. Better if you can use some statistics to get more reliable recognition.<br />Andyreyhttps://www.blogger.com/profile/09146432959603914745noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-8427493418250921892022-05-12T08:23:13.122-04:002022-05-12T08:23:13.122-04:00Hello, If I want to add more faces to the library ...Hello, If I want to add more faces to the library would it be necessary to retrain it? Do you think this library would be accurate to compare faces from a security camera frame with let's say a prison database picture? Or should I do some pre treatment on the frame to enhance the image quality? Thanks in advance.Anonymoushttps://www.blogger.com/profile/04493212271012455282noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-28305459691609731272022-02-21T09:05:52.994-05:002022-02-21T09:05:52.994-05:00Hi Davis, okay thank you for your quick response, ...Hi Davis, okay thank you for your quick response, thanks again for your great work in dlib.junehttps://www.blogger.com/profile/16265435138759586833noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-63568702246220717842022-02-21T08:36:16.421-05:002022-02-21T08:36:16.421-05:00No, but the test can be rerun using http://dlib.ne...No, but the test can be rerun using http://dlib.net/files/dlib_face_recognition_resnet_model_v1_lfw_test_scripts.tar.bz2, reproducing the dlib results noted here.Davis Kinghttps://www.blogger.com/profile/16577392965630448489noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-78316872657947758892022-02-21T03:20:20.118-05:002022-02-21T03:20:20.118-05:00Hello, Does dlib recognition have a paper on the R...Hello, Does dlib recognition have a paper on the ROC curve on the LFW benchmark and prediction times, same on this paper for comparison to other techniques. https://www.cs.cmu.edu/~satya/docdir/CMU-CS-16-118.pdf <br /><br />Thank youjunehttps://www.blogger.com/profile/16265435138759586833noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-8014088940923167862021-10-30T10:07:40.798-04:002021-10-30T10:07:40.798-04:00Er, I forgot what the values are. The default mar...Er, I forgot what the values are. The default margin is 0.04 instead of 1.0. But that numerical detail doesn't matter. It's a max margin loss for any number > 0 in that part of the equation.Davis Kinghttps://www.blogger.com/profile/16577392965630448489noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-68814654587019317892021-10-30T10:05:31.186-04:002021-10-30T10:05:31.186-04:00It's just a classic hinge loss. https://en.wi...It's just a classic hinge loss. https://en.wikipedia.org/wiki/Hinge_loss. Except with 0.6 instead of 1.0. The value there is smaller because it worked better with typical SGD and weight decay settings, which have a hard time making larger network outputs.<br /><br />Random variations can't push it over a threshold due to the margin not being 0 in the loss. You aren't reading the loss function right since if I'm understanding you you are saying the margin looks like it's 0. It's not 0. This is a class max-margin style loss.Davis Kinghttps://www.blogger.com/profile/16577392965630448489noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-9384506141597157172021-10-20T09:01:47.833-04:002021-10-20T09:01:47.833-04:00I've looked at the metric_loss code and I wond...I've looked at the metric_loss code and I wonder why the loss function is defined so "softly" so to speak. Any distance less than 0.6 will result in zero loss for members of the same class, the same for any distance above 0.6 for members of different classes. There is no incentive for the network to make a more compact representation for each class, so even if the network is trained "perfectly", any random variations can push the result over the threshold. <br />Would it be a good idea to have a separate distance threshold for positive cases (for example 0.3), lower than the distance threshold for negative cases?anszomhttps://www.blogger.com/profile/14960203120276718789noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-2039368204286062462020-11-09T15:03:45.045-05:002020-11-09T15:03:45.045-05:00This comment has been removed by the author.Victorhttps://www.blogger.com/profile/13616391485068779800noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-67725329469958328522020-11-09T15:02:53.895-05:002020-11-09T15:02:53.895-05:00Hi Davis, fist I want to say Thanks so much for yo...Hi Davis, fist I want to say Thanks so much for your dlib library, help me a lot for do my scripts, Also I want to ask You about the http://dlib.net/files/dlib_face_recognition_resnet_model_v1_lfw_test_scripts.tar.bz2 I’m trying to run this test doing your read.txt and trying to run the ./main, but I don’t getting the rest of the results, it’s only cout the disth thresh: 0,6 and margin: 0.04<br />And don’t show anything else, it’s like the code just stop running... please help me to finish this! Thanks a lot Victorhttps://www.blogger.com/profile/13616391485068779800noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-76484055733460374972020-08-02T07:46:10.364-04:002020-08-02T07:46:10.364-04:00There isn't any special stuff for ARM in dlib&...There isn't any special stuff for ARM in dlib's DNN tooling. The DNN tooling is really targeted at platforms that have cuDNN. Everything else will be fairly slow.Davis Kinghttps://www.blogger.com/profile/16577392965630448489noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-45103011777282401512020-07-31T06:54:53.034-04:002020-07-31T06:54:53.034-04:00I have profiled the dlib facial recognition exampl...I have profiled the dlib facial recognition example with perf. There are 3 hot spots:<br />´´´<br /> 49.03% A20_Face A20_Face [.] sgemm_kernel_L4_M4_22 <br /> 12.75% A20_Face A20_Face [.] sgemm_ncopy_L4_M4_20 <br /> 11.96% A20_Face A20_Face [.] dlib::cpu::img2col <br />´´´<br />I believe the first 2 functions are covered by OpenBLAS, the third is a dlib function.<br />Any recommendation regarding speedup on ARM CPU?Mikehttps://www.blogger.com/profile/15549468676645780411noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-45980405695658841832020-07-30T07:42:58.146-04:002020-07-30T07:42:58.146-04:00No current plans to expand to other BLAS like APIs...No current plans to expand to other BLAS like APIs. Note however that dlib will already work with any BLAS and LAPACK libraries that provide the standard CBLAS API and typical LAPACK link symbols.Davis Kinghttps://www.blogger.com/profile/16577392965630448489noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-72455519959045474752020-07-29T09:36:41.654-04:002020-07-29T09:36:41.654-04:00Hi Davis,
do you have any plans to support "b...Hi Davis,<br />do you have any plans to support "blasfeo" instead of OpenBLAS?Mikehttps://www.blogger.com/profile/15549468676645780411noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-59831248240224648402020-07-29T04:51:30.901-04:002020-07-29T04:51:30.901-04:00As far as training was held on general datasets by...As far as training was held on general datasets by Davis, I suppose, there is a bias to europium (Caucasoid ) type, and I would specify the question:<br /> - if I want to distinct among special group of people, such as Asian, Black, or children, is there a tool for fine tuning from the original dlib model? This is very actual question for many. Andyreyhttps://www.blogger.com/profile/09146432959603914745noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-25902881388242825132020-07-29T02:36:52.030-04:002020-07-29T02:36:52.030-04:00i am using make app face matching . but does not ...i am using make app face matching . but does not work in japan ?<br />Anonymoushttps://www.blogger.com/profile/08610260001876465286noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-51423989497639439792020-07-29T02:35:20.536-04:002020-07-29T02:35:20.536-04:00can i use this dlib library in japan country.?can i use this dlib library in japan country.?Anonymoushttps://www.blogger.com/profile/08610260001876465286noreply@blogger.comtag:blogger.com,1999:blog-6061887630060661987.post-77526305923395604132020-07-21T10:21:09.503-04:002020-07-21T10:21:09.503-04:00Sir i didnt find and documentations. please help m...Sir i didnt find and documentations. please help me Mahesh Bishthttps://www.blogger.com/profile/09296643362240018338noreply@blogger.com