how to realise face alignment?

[weiyt2014]

Recently,I have been read your code!The code has realise 68 feature landmarks alignment,but the whole face don't follow the 68 feature landmarks aligement to realise face alignment.How to realise it?And I found rotateshape function,And how to use the rotate function in the code? I found the globalprediction.m function don't have the rotateshape function,Whether to need the function?And how to use it? I need some help!Thanks very much!

[jwyang]

@weiyt2014 , I am sorry that I do not understand your first question. As for the second question, rotateshape function is used for generate new training samples in the training stage, which need not to be used in the testing stage.

[weiyt2014]

My first problem is showed in the image which is obtained by test.m function!The image has the red¡¢blue¡¢green landmark points.68 red landmark points are obtained by man-made before face alignment. 68 blue landmark or 68 green landmark points are obtained by train model.The locations of points is different from the red points,but the face images don't have changed.I want to express the face images should also change and should also be the same as the change of the blue or green landmark points.I want to ask how to realise it? I hope to obtain some suggestions! Thank verymuch! I send you a image!

At 2014-11-18 14:56:08, "jwyang" [email protected] wrote:

@weiyt2014 , I am sorry that I do not understand your first question. As for the second question, rotateshape function is used for generate new training samples in the training stage, which need not to be used in the testing stage.

¡ª Reply to this email directly or view it on GitHub.

[jwyang]

You mean you want to do some image transformation given the landmarks? One method is to do triangulation according to the landmarks, and then conduct affine transformation for each triangular to the specified points, say frontal face. For affine transformation, please refer to http://docs.opencv.org/doc/tutorials/imgproc/imgtrans/warp_affine/warp_affine.html

At 2014-11-18 17:08:41, "weiyt2014" [email protected] wrote:

My first problem is showed in the image which is obtained by test.m function!The image has the red¡¢blue¡¢green landmark points.68 red landmark points are obtained by man-made before face alignment. 68 blue landmark or 68 green landmark points are obtained by train model.The locations of points is different from the red points,but the face images don't have changed.I want to express the face images should also change and should also be the same as the change of the blue or green landmark points.I want to ask how to realise it? I hope to obtain some suggestions! Thank verymuch! I send you a image!

At 2014-11-18 14:56:08, "jwyang" [email protected] wrote:

@weiyt2014 , I am sorry that I do not understand your first question. As for the second question, rotateshape function is used for generate new training samples in the training stage, which need not to be used in the testing stage.

¡ª Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

[weiyt2014]

Thanks very much!Yes,My meaning is so.Do your code realise the effect?

At 2014-11-18 23:45:37, "jwyang" [email protected] wrote: You mean you want to do some image transformation given the landmarks? One method is to do triangulation according to the landmarks, and then conduct affine transformation for each triangular to the specified points, say frontal face. For affine transformation, please refer to http://docs.opencv.org/doc/tutorials/imgproc/imgtrans/warp_affine/warp_affine.html

At 2014-11-18 17:08:41, "weiyt2014" [email protected] wrote:

My first problem is showed in the image which is obtained by test.m function!The image has the red¡¢blue¡¢green landmark points.68 red landmark points are obtained by man-made before face alignment. 68 blue landmark or 68 green landmark points are obtained by train model.The locations of points is different from the red points,but the face images don't have changed.I want to express the face images should also change and should also be the same as the change of the blue or green landmark points.I want to ask how to realise it? I hope to obtain some suggestions! Thank verymuch! I send you a image!

At 2014-11-18 14:56:08, "jwyang" [email protected] wrote:

@weiyt2014 , I am sorry that I do not understand your first question. As for the second question, rotateshape function is used for generate new training samples in the training stage, which need not to be used in the testing stage.

¡ª Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

[jwyang]

Sorry, I did not do this.

At 2014-11-19 08:06:07, "weiyt2014" [email protected] wrote:

Thanks very much!Yes,My meaning is so.Do your code realise the effect?

At 2014-11-18 23:45:37, "jwyang" [email protected] wrote: You mean you want to do some image transformation given the landmarks? One method is to do triangulation according to the landmarks, and then conduct affine transformation for each triangular to the specified points, say frontal face. For affine transformation, please refer to http://docs.opencv.org/doc/tutorials/imgproc/imgtrans/warp_affine/warp_affine.html

At 2014-11-18 17:08:41, "weiyt2014" [email protected] wrote:

My first problem is showed in the image which is obtained by test.m function!The image has the red¡¢blue¡¢green landmark points.68 red landmark points are obtained by man-made before face alignment. 68 blue landmark or 68 green landmark points are obtained by train model.The locations of points is different from the red points,but the face images don't have changed.I want to express the face images should also change and should also be the same as the change of the blue or green landmark points.I want to ask how to realise it? I hope to obtain some suggestions! Thank verymuch! I send you a image!

At 2014-11-18 14:56:08, "jwyang" [email protected] wrote:

@weiyt2014 , I am sorry that I do not understand your first question. As for the second question, rotateshape function is used for generate new training samples in the training stage, which need not to be used in the testing stage.

¡ª Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

[weiyt2014]

I'm sorry to interrupt you! I hope to know how to produce*.pts file? And how to obtain 68 landmarks? And are they obtained by face dection?

At 2014-11-19 13:31:26, "jwyang" [email protected] wrote: Sorry, I did not do this.

At 2014-11-19 08:06:07, "weiyt2014" [email protected] wrote:

Thanks very much!Yes,My meaning is so.Do your code realise the effect?

At 2014-11-18 23:45:37, "jwyang" [email protected] wrote: You mean you want to do some image transformation given the landmarks? One method is to do triangulation according to the landmarks, and then conduct affine transformation for each triangular to the specified points, say frontal face. For affine transformation, please refer to http://docs.opencv.org/doc/tutorials/imgproc/imgtrans/warp_affine/warp_affine.html

At 2014-11-18 17:08:41, "weiyt2014" [email protected] wrote:

My first problem is showed in the image which is obtained by test.m function!The image has the red¡¢blue¡¢green landmark points.68 red landmark points are obtained by man-made before face alignment. 68 blue landmark or 68 green landmark points are obtained by train model.The locations of points is different from the red points,but the face images don't have changed.I want to express the face images should also change and should also be the same as the change of the blue or green landmark points.I want to ask how to realise it? I hope to obtain some suggestions! Thank verymuch! I send you a image!

At 2014-11-18 14:56:08, "jwyang" [email protected] wrote:

@weiyt2014 , I am sorry that I do not understand your first question. As for the second question, rotateshape function is used for generate new training samples in the training stage, which need not to be used in the testing stage.

¡ª Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

[jwyang]

Hi, you can refer to the format of *.pts file in the dataset, e.g. lfpw. To obtain 68 points, you may need to modify the config_tr.m and config_te.m. Find command: params.ind_usedpts = 18:68; and change it to params.ind_usedpts = 1:68; Then, retrain the model.

At 2014-11-21 08:09:10, "weiyt2014" [email protected] wrote: I'm sorry to interrupt you! I hope to know how to produce*.pts file? And how to obtain 68 landmarks? And are they obtained by face dection?

At 2014-11-19 13:31:26, "jwyang" [email protected] wrote: Sorry, I did not do this.

At 2014-11-19 08:06:07, "weiyt2014" [email protected] wrote:

Thanks very much!Yes,My meaning is so.Do your code realise the effect?

At 2014-11-18 23:45:37, "jwyang" [email protected] wrote: You mean you want to do some image transformation given the landmarks? One method is to do triangulation according to the landmarks, and then conduct affine transformation for each triangular to the specified points, say frontal face. For affine transformation, please refer to http://docs.opencv.org/doc/tutorials/imgproc/imgtrans/warp_affine/warp_affine.html

At 2014-11-18 17:08:41, "weiyt2014" [email protected] wrote:

My first problem is showed in the image which is obtained by test.m function!The image has the red¡¢blue¡¢green landmark points.68 red landmark points are obtained by man-made before face alignment. 68 blue landmark or 68 green landmark points are obtained by train model.The locations of points is different from the red points,but the face images don't have changed.I want to express the face images should also change and should also be the same as the change of the blue or green landmark points.I want to ask how to realise it? I hope to obtain some suggestions! Thank verymuch! I send you a image!

At 2014-11-18 14:56:08, "jwyang" [email protected] wrote:

@weiyt2014 , I am sorry that I do not understand your first question. As for the second question, rotateshape function is used for generate new training samples in the training stage, which need not to be used in the testing stage.

¡ª Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

[weiyt2014]

I'm sorry to interrupt to you again! I have a problem to need your suggestions! I hope to learn what function *.pts file has in test.m! I feel *.pts file is needed in the train process, but I feel it may give up in test process! whether *.pts is to determine a shape region in loadsample.m? If I hope to get the region from *.jpg or *.png, how can I realise it? I hope to get some suggestions!Thanks very much!

At 2014-11-18 23:45:37, "jwyang" [email protected] wrote: You mean you want to do some image transformation given the landmarks? One method is to do triangulation according to the landmarks, and then conduct affine transformation for each triangular to the specified points, say frontal face. For affine transformation, please refer to http://docs.opencv.org/doc/tutorials/imgproc/imgtrans/warp_affine/warp_affine.html

At 2014-11-18 17:08:41, "weiyt2014" [email protected] wrote:

My first problem is showed in the image which is obtained by test.m function!The image has the red¡¢blue¡¢green landmark points.68 red landmark points are obtained by man-made before face alignment. 68 blue landmark or 68 green landmark points are obtained by train model.The locations of points is different from the red points,but the face images don't have changed.I want to express the face images should also change and should also be the same as the change of the blue or green landmark points.I want to ask how to realise it? I hope to obtain some suggestions! Thank verymuch! I send you a image!

At 2014-11-18 14:56:08, "jwyang" [email protected] wrote:

@weiyt2014 , I am sorry that I do not understand your first question. As for the second question, rotateshape function is used for generate new training samples in the training stage, which need not to be used in the testing stage.

¡ª Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

[jwyang]

Hi, the pts file in the test procedure is used for groundtruth for the evaluation of perfomance. Not only it is used to obtain face region, but also it is used to compute the face alignment error on landmarks.

At 2014-11-28 08:35:50, "weiyt2014" [email protected] wrote:

I'm sorry to interrupt to you again! I have a problem to need your suggestions! I hope to learn what function *.pts file has in test.m! I feel *.pts file is needed in the train process, but I feel it may give up in test process! whether *.pts is to determine a shape region in loadsample.m? If I hope to get the region from *.jpg or *.png, how can I realise it? I hope to get some suggestions!Thanks very much!

At 2014-11-18 23:45:37, "jwyang" [email protected] wrote: You mean you want to do some image transformation given the landmarks? One method is to do triangulation according to the landmarks, and then conduct affine transformation for each triangular to the specified points, say frontal face. For affine transformation, please refer to http://docs.opencv.org/doc/tutorials/imgproc/imgtrans/warp_affine/warp_affine.html

At 2014-11-18 17:08:41, "weiyt2014" [email protected] wrote:

My first problem is showed in the image which is obtained by test.m function!The image has the red¡¢blue¡¢green landmark points.68 red landmark points are obtained by man-made before face alignment. 68 blue landmark or 68 green landmark points are obtained by train model.The locations of points is different from the red points,but the face images don't have changed.I want to express the face images should also change and should also be the same as the change of the blue or green landmark points.I want to ask how to realise it? I hope to obtain some suggestions! Thank verymuch! I send you a image!

At 2014-11-18 14:56:08, "jwyang" [email protected] wrote:

@weiyt2014 , I am sorry that I do not understand your first question. As for the second question, rotateshape function is used for generate new training samples in the training stage, which need not to be used in the testing stage.

¡ª Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

[weiyt2014]

how to obtain *.pts file ? Can it be replaced by *.jpg? Because we often obtain *.jpg image and don't have *.pts!

At 2014-11-28 09:42:53, "jwyang" [email protected] wrote: Hi, the pts file in the test procedure is used for groundtruth for the evaluation of perfomance. Not only it is used to obtain face region, but also it is used to compute the face alignment error on landmarks.

At 2014-11-28 08:35:50, "weiyt2014" [email protected] wrote:

I'm sorry to interrupt to you again! I have a problem to need your suggestions! I hope to learn what function *.pts file has in test.m! I feel *.pts file is needed in the train process, but I feel it may give up in test process! whether *.pts is to determine a shape region in loadsample.m? If I hope to get the region from *.jpg or *.png, how can I realise it? I hope to get some suggestions!Thanks very much!

At 2014-11-18 23:45:37, "jwyang" [email protected] wrote: You mean you want to do some image transformation given the landmarks? One method is to do triangulation according to the landmarks, and then conduct affine transformation for each triangular to the specified points, say frontal face. For affine transformation, please refer to http://docs.opencv.org/doc/tutorials/imgproc/imgtrans/warp_affine/warp_affine.html

At 2014-11-18 17:08:41, "weiyt2014" [email protected] wrote:

My first problem is showed in the image which is obtained by test.m function!The image has the red¡¢blue¡¢green landmark points.68 red landmark points are obtained by man-made before face alignment. 68 blue landmark or 68 green landmark points are obtained by train model.The locations of points is different from the red points,but the face images don't have changed.I want to express the face images should also change and should also be the same as the change of the blue or green landmark points.I want to ask how to realise it? I hope to obtain some suggestions! Thank verymuch! I send you a image!

At 2014-11-18 14:56:08, "jwyang" [email protected] wrote:

@weiyt2014 , I am sorry that I do not understand your first question. As for the second question, rotateshape function is used for generate new training samples in the training stage, which need not to be used in the testing stage.

¡ª Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

[jwyang]

Actually, if you just want to get the landmarks for testing image, you just comment those codes for *.pts.

At 2014-11-28 10:38:09, "weiyt2014" [email protected] wrote: how to obtain *.pts file ? Can it be replaced by *.jpg? Because we often obtain *.jpg image and don't have *.pts!

At 2014-11-28 09:42:53, "jwyang" [email protected] wrote: Hi, the pts file in the test procedure is used for groundtruth for the evaluation of perfomance. Not only it is used to obtain face region, but also it is used to compute the face alignment error on landmarks.

At 2014-11-28 08:35:50, "weiyt2014" [email protected] wrote:

I'm sorry to interrupt to you again! I have a problem to need your suggestions! I hope to learn what function *.pts file has in test.m! I feel *.pts file is needed in the train process, but I feel it may give up in test process! whether *.pts is to determine a shape region in loadsample.m? If I hope to get the region from *.jpg or *.png, how can I realise it? I hope to get some suggestions!Thanks very much!

At 2014-11-18 23:45:37, "jwyang" [email protected] wrote: You mean you want to do some image transformation given the landmarks? One method is to do triangulation according to the landmarks, and then conduct affine transformation for each triangular to the specified points, say frontal face. For affine transformation, please refer to http://docs.opencv.org/doc/tutorials/imgproc/imgtrans/warp_affine/warp_affine.html

At 2014-11-18 17:08:41, "weiyt2014" [email protected] wrote:

My first problem is showed in the image which is obtained by test.m function!The image has the red¡¢blue¡¢green landmark points.68 red landmark points are obtained by man-made before face alignment. 68 blue landmark or 68 green landmark points are obtained by train model.The locations of points is different from the red points,but the face images don't have changed.I want to express the face images should also change and should also be the same as the change of the blue or green landmark points.I want to ask how to realise it? I hope to obtain some suggestions! Thank verymuch! I send you a image!

At 2014-11-18 14:56:08, "jwyang" [email protected] wrote:

@weiyt2014 , I am sorry that I do not understand your first question. As for the second question, rotateshape function is used for generate new training samples in the training stage, which need not to be used in the testing stage.

¡ª Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

[weiyt2014]

Sorry to interrupt you!LBFRegmodel includes LBFRegModel.randf and LBFRegModel and Ws that are 7_1 cell. I hope to know Why randf is 68_10,and Ws is 10880_136? Especially 10880_136 of Ws show what?

At 2014-11-28 18:27:09, "jwyang" [email protected] wrote: Actually, if you just want to get the landmarks for testing image, you just comment those codes for *.pts.

At 2014-11-28 10:38:09, "weiyt2014" [email protected] wrote: how to obtain *.pts file ? Can it be replaced by *.jpg? Because we often obtain *.jpg image and don't have *.pts!

At 2014-11-28 09:42:53, "jwyang" [email protected] wrote: Hi, the pts file in the test procedure is used for groundtruth for the evaluation of perfomance. Not only it is used to obtain face region, but also it is used to compute the face alignment error on landmarks.

At 2014-11-28 08:35:50, "weiyt2014" [email protected] wrote:

I'm sorry to interrupt to you again! I have a problem to need your suggestions! I hope to learn what function *.pts file has in test.m! I feel *.pts file is needed in the train process, but I feel it may give up in test process! whether *.pts is to determine a shape region in loadsample.m? If I hope to get the region from *.jpg or *.png, how can I realise it? I hope to get some suggestions!Thanks very much!

At 2014-11-18 23:45:37, "jwyang" [email protected] wrote: You mean you want to do some image transformation given the landmarks? One method is to do triangulation according to the landmarks, and then conduct affine transformation for each triangular to the specified points, say frontal face. For affine transformation, please refer to http://docs.opencv.org/doc/tutorials/imgproc/imgtrans/warp_affine/warp_affine.html

At 2014-11-18 17:08:41, "weiyt2014" [email protected] wrote:

My first problem is showed in the image which is obtained by test.m function!The image has the red¡¢blue¡¢green landmark points.68 red landmark points are obtained by man-made before face alignment. 68 blue landmark or 68 green landmark points are obtained by train model.The locations of points is different from the red points,but the face images don't have changed.I want to express the face images should also change and should also be the same as the change of the blue or green landmark points.I want to ask how to realise it? I hope to obtain some suggestions! Thank verymuch! I send you a image!

At 2014-11-18 14:56:08, "jwyang" [email protected] wrote:

@weiyt2014 , I am sorry that I do not understand your first question. As for the second question, rotateshape function is used for generate new training samples in the training stage, which need not to be used in the testing stage.

¡ª Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

[jwyang]

We have 68 landmarks, and each landmark corresponds to 10 random trees. Therefore, we will have 68 * 10 random trees overall. Each random tree has 2^4 leaf nodes, so the dimension of local binary feature is 16 * 68 * 10. Therefore the Ws is a 10880 * 136 matrix, whose dimension need to be consistent to the dimension of local binary feature.

On Thu, Dec 4, 2014 at 2:49 AM, weiyt2014 [email protected] wrote:

Sorry to interrupt you!LBFRegmodel includes LBFRegModel.randf and LBFRegModel and Ws that are 7_1 cell. I hope to know Why randf is 68_10,and Ws is 10880_136? Especially 10880_136 of Ws show what?

At 2014-11-28 18:27:09, "jwyang" [email protected] wrote: Actually, if you just want to get the landmarks for testing image, you just comment those codes for *.pts.

At 2014-11-28 10:38:09, "weiyt2014" [email protected] wrote: how to obtain *.pts file ? Can it be replaced by *.jpg? Because we often obtain *.jpg image and don't have *.pts!

At 2014-11-28 09:42:53, "jwyang" [email protected] wrote: Hi, the pts file in the test procedure is used for groundtruth for the evaluation of perfomance. Not only it is used to obtain face region, but also it is used to compute the face alignment error on landmarks.

At 2014-11-28 08:35:50, "weiyt2014" [email protected] wrote:

I'm sorry to interrupt to you again! I have a problem to need your suggestions! I hope to learn what function *.pts file has in test.m! I feel *.pts file is needed in the train process, but I feel it may give up in test process! whether *.pts is to determine a shape region in loadsample.m? If I hope to get the region from *.jpg or *.png, how can I realise it? I hope to get some suggestions!Thanks very much!

At 2014-11-18 23:45:37, "jwyang" [email protected] wrote: You mean you want to do some image transformation given the landmarks? One method is to do triangulation according to the landmarks, and then conduct affine transformation for each triangular to the specified points, say frontal face. For affine transformation, please refer to http://docs.opencv.org/doc/tutorials/imgproc/imgtrans/warp_affine/warp_affine.html

At 2014-11-18 17:08:41, "weiyt2014" [email protected] wrote:

My first problem is showed in the image which is obtained by test.m function!The image has the red¡¢blue¡¢green landmark points.68 red landmark points are obtained by man-made before face alignment. 68 blue landmark or 68 green landmark points are obtained by train model.The locations of points is different from the red points,but the face images don't have changed.I want to express the face images should also change and should also be the same as the change of the blue or green landmark points.I want to ask how to realise it? I hope to obtain some suggestions! Thank verymuch! I send you a image!

At 2014-11-18 14:56:08, "jwyang" [email protected] wrote:

@weiyt2014 , I am sorry that I do not understand your first question. As for the second question, rotateshape function is used for generate new training samples in the training stage, which need not to be used in the testing stage.

¡ª Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub https://github.com/jwyang/face-alignment/issues/6#issuecomment-65553541.

Jianwei Yang Beijing University of Posts and Telecommunications Institute of Automation, Chinese Academy of Sciences 14th floor in Intelligence Building 95 Zhongguancun East Road Haidian District, Beijing 100190, China

杨健伟 北京邮电大学 中国科学院自动化研究所 北京海淀区中关村东路95号智能化大厦14层 邮政编码：100190

[weiyt2014]

Professor Yang, I get "Data{i}.bbox_gt" by Face detection and I get a result!The result is like Figure1.I found the 68 landmarks are different from expected results.I found the question may be caused of globalprediction.m and drawshapes.m. If 68 landmarks are not given,"Te_Data{k}.shape_gt" cannot be obtained,and "drawshapes(Te_Data{k}.img_gray, [Te_Data{k}.shape_gt Te_Data{k}.intermediate_shapes{1}(:,:, s) shape_newstage])" will not be executed.I have used "meanshape_resize" instead of "Te_Data{k}.shape_gt", and the result is showed in Figure 1.I found " plot(shapes(is, 2_(s-1) + 1), shapes(is, 2_s), colors{s}, 'LineWidth', 1)" in drawshapes.m, I found 2_(s-1) and 2_s have effects on the location of 68 landmarks,but I don't know how to change it,I hope to get some suggestions! Thanks very much!

At 2014-12-05 05:08:57, "jwyang" [email protected] wrote: We have 68 landmarks, and each landmark corresponds to 10 random trees. Therefore, we will have 68 * 10 random trees overall. Each random tree has 2^4 leaf nodes, so the dimension of local binary feature is 16 * 68 * 10. Therefore the Ws is a 10880 * 136 matrix, whose dimension need to be consistent to the dimension of local binary feature.

On Thu, Dec 4, 2014 at 2:49 AM, weiyt2014 [email protected] wrote:

Sorry to interrupt you!LBFRegmodel includes LBFRegModel.randf and LBFRegModel and Ws that are 7_1 cell. I hope to know Why randf is 68_10,and Ws is 10880_136? Especially 10880_136 of Ws show what?

At 2014-11-28 18:27:09, "jwyang" [email protected] wrote: Actually, if you just want to get the landmarks for testing image, you just comment those codes for *.pts.

At 2014-11-28 10:38:09, "weiyt2014" [email protected] wrote: how to obtain *.pts file ? Can it be replaced by *.jpg? Because we often obtain *.jpg image and don't have *.pts!

At 2014-11-28 09:42:53, "jwyang" [email protected] wrote: Hi, the pts file in the test procedure is used for groundtruth for the evaluation of perfomance. Not only it is used to obtain face region, but also it is used to compute the face alignment error on landmarks.

At 2014-11-28 08:35:50, "weiyt2014" [email protected] wrote:

I'm sorry to interrupt to you again! I have a problem to need your suggestions! I hope to learn what function *.pts file has in test.m! I feel *.pts file is needed in the train process, but I feel it may give up in test process! whether *.pts is to determine a shape region in loadsample.m? If I hope to get the region from *.jpg or *.png, how can I realise it? I hope to get some suggestions!Thanks very much!

At 2014-11-18 23:45:37, "jwyang" [email protected] wrote: You mean you want to do some image transformation given the landmarks? One method is to do triangulation according to the landmarks, and then conduct affine transformation for each triangular to the specified points, say frontal face. For affine transformation, please refer to http://docs.opencv.org/doc/tutorials/imgproc/imgtrans/warp_affine/warp_affine.html

At 2014-11-18 17:08:41, "weiyt2014" [email protected] wrote:

My first problem is showed in the image which is obtained by test.m function!The image has the red¡¢blue¡¢green landmark points.68 red landmark points are obtained by man-made before face alignment. 68 blue landmark or 68 green landmark points are obtained by train model.The locations of points is different from the red points,but the face images don't have changed.I want to express the face images should also change and should also be the same as the change of the blue or green landmark points.I want to ask how to realise it? I hope to obtain some suggestions! Thank verymuch! I send you a image!

At 2014-11-18 14:56:08, "jwyang" [email protected] wrote:

@weiyt2014 , I am sorry that I do not understand your first question. As for the second question, rotateshape function is used for generate new training samples in the training stage, which need not to be used in the testing stage.

¡ª Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub https://github.com/jwyang/face-alignment/issues/6#issuecomment-65553541.

Jianwei Yang Beijing University of Posts and Telecommunications Institute of Automation, Chinese Academy of Sciences 14th floor in Intelligence Building 95 Zhongguancun East Road Haidian District, Beijing 100190, China

杨健伟 北京邮电大学 中国科学院自动化研究所 北京海淀区中关村东路95号智能化大厦14层 邮政编码：100190

— Reply to this email directly or view it on GitHub.

[jwyang]

Hi weiyt, I am sorry that I cannot see your figures. But i guess you mean that there is something wrong when you replace Te_Data{k}.shape_gt with meanshaoe_resize? In the drawshapes.m, 2*(s-1) + 1 and 2 * s are the x and y coordinates of s-th shape, respectively. You do not have to change it. What you should do is changing drawshapes(Te_Data{k}.img_gray, [Te_Data{k}.shape_gt Te_Data{k}.intermediate_shapes{1}(:,:, s) shape_newstage]), ensure that each shape is a 68 X 2 matrix.

On Fri, Dec 5, 2014 at 4:48 AM, weiyt2014 [email protected] wrote:

Professor Yang, I get "Data{i}.bbox_gt" by Face detection and I get a result!The result is like Figure1.I found the 68 landmarks are different from expected results.I found the question may be caused of globalprediction.m and drawshapes.m. If 68 landmarks are not given,"Te_Data{k}.shape_gt" cannot be obtained,and "drawshapes(Te_Data{k}.img_gray, [Te_Data{k}.shape_gt Te_Data{k}.intermediate_shapes{1}(:,:, s) shape_newstage])" will not be executed.I have used "meanshape_resize" instead of "Te_Data{k}.shape_gt", and the result is showed in Figure 1.I found " plot(shapes(is, 2_(s-1) + 1), shapes(is, 2_s), colors{s}, 'LineWidth', 1)" in drawshapes.m, I found 2_(s-1) and 2_s have effects on the location of 68 landmarks,but I don't know how to change it,I hope to get some suggestions! Thanks very much!

At 2014-12-05 05:08:57, "jwyang" [email protected] wrote: We have 68 landmarks, and each landmark corresponds to 10 random trees. Therefore, we will have 68 * 10 random trees overall. Each random tree has 2^4 leaf nodes, so the dimension of local binary feature is 16 * 68 * 10. Therefore the Ws is a 10880 * 136 matrix, whose dimension need to be consistent to the dimension of local binary feature.

On Thu, Dec 4, 2014 at 2:49 AM, weiyt2014 [email protected] wrote:

Sorry to interrupt you!LBFRegmodel includes LBFRegModel.randf and LBFRegModel and Ws that are 7_1 cell. I hope to know Why randf is 68_10,and Ws is 10880_136? Especially 10880_136 of Ws show what?

At 2014-11-28 18:27:09, "jwyang" [email protected] wrote: Actually, if you just want to get the landmarks for testing image, you just comment those codes for *.pts.

At 2014-11-28 10:38:09, "weiyt2014" [email protected] wrote: how to obtain *.pts file ? Can it be replaced by *.jpg? Because we often obtain *.jpg image and don't have *.pts!

At 2014-11-28 09:42:53, "jwyang" [email protected] wrote: Hi, the pts file in the test procedure is used for groundtruth for the evaluation of perfomance. Not only it is used to obtain face region, but also it is used to compute the face alignment error on landmarks.

At 2014-11-28 08:35:50, "weiyt2014" [email protected] wrote:

I'm sorry to interrupt to you again! I have a problem to need your suggestions! I hope to learn what function *.pts file has in test.m! I feel *.pts file is needed in the train process, but I feel it may give up in test process! whether *.pts is to determine a shape region in loadsample.m? If I hope to get the region from *.jpg or *.png, how can I realise it? I hope to get some suggestions!Thanks very much!

At 2014-11-18 23:45:37, "jwyang" [email protected] wrote: You mean you want to do some image transformation given the landmarks? One method is to do triangulation according to the landmarks, and then conduct affine transformation for each triangular to the specified points, say frontal face. For affine transformation, please refer to

http://docs.opencv.org/doc/tutorials/imgproc/imgtrans/warp_affine/warp_affine.html

At 2014-11-18 17:08:41, "weiyt2014" [email protected] wrote:

My first problem is showed in the image which is obtained by test.m function!The image has the red¡¢blue¡¢green landmark points.68 red landmark points are obtained by man-made before face alignment. 68 blue landmark or 68 green landmark points are obtained by train model.The locations of points is different from the red points,but the face images don't have changed.I want to express the face images should also change and should also be the same as the change of the blue or green landmark points.I want to ask how to realise it? I hope to obtain some suggestions! Thank verymuch! I send you a image!

At 2014-11-18 14:56:08, "jwyang" [email protected] wrote:

@weiyt2014 , I am sorry that I do not understand your first question. As for the second question, rotateshape function is used for generate new training samples in the training stage, which need not to be used in the testing stage.

¡ª Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub <https://github.com/jwyang/face-alignment/issues/6#issuecomment-65553541 .

Jianwei Yang Beijing University of Posts and Telecommunications Institute of Automation, Chinese Academy of Sciences 14th floor in Intelligence Building 95 Zhongguancun East Road Haidian District, Beijing 100190, China

杨健伟 北京邮电大学 中国科学院自动化研究所 北京海淀区中关村东路95号智能化大厦14层 邮政编码：100190

— Reply to this email directly or view it on GitHub.

— Reply to this email directly or view it on GitHub https://github.com/jwyang/face-alignment/issues/6#issuecomment-65773902.

Jianwei Yang Beijing University of Posts and Telecommunications Institute of Automation, Chinese Academy of Sciences 14th floor in Intelligence Building 95 Zhongguancun East Road Haidian District, Beijing 100190, China

杨健伟 北京邮电大学 中国科学院自动化研究所 北京海淀区中关村东路95号智能化大厦14层 邮政编码：100190