search

FunkyKoki / Look_At_Boundary_PyTorch

A PyTorch re-implement of CVPR 2018 LAB(Look At Bounday) --Champagne Jin
9 stars 1 forks source link

Hi！ how to draw boundary heat map? #6

Open wqz960 opened 5 years ago

wqz960 commented 5 years ago

Great job! But I want to get a boundary heat map by inputing a raw image(cropped) and groundtrue landmarks, but I saw your code, i cannot understand.... can you provide me a script for drawing boundary heat map by inputing the image and landmarks instead of dataset in your get_gt_heatmap(dataset, gt_coords)function or tell me more details to change it. Thank you!

wqz960 commented 5 years ago

Thanks to your codes. I have drawn the gt heat maps. but there is another question, the size of heat maps is (13,64,64), the pixel value is [0,1], but the pixel value of raw image from the opencv imreader is [0,255] , does it need to transform the heat maps to [0,255]?

FunkyKoki commented 5 years ago

It is great if it helps you. About the question you asked, use the matplotlib, it can draw heat map by pixel in range(0, 1).

---Original--- From: "wqz960"notifications@github.com Date: Sat, Jun 8, 2019 15:31 PM To: "FunkyKoki/Look_At_Boundary_PyTorch"Look_At_Boundary_PyTorch@noreply.github.com; Cc: "Subscribed"subscribed@noreply.github.com; Subject: Re: [FunkyKoki/Look_At_Boundary_PyTorch] Hi！ how to draw boundary heat map? (#6)

Thanks to your codes. I have drawn the gt heat maps. but there is another question, the size of heat maps is (13,64,64), the pixel value is [0,1], but the pixel value of raw image from the opencv imreader is [0,255] , does it need to transform the heat maps to [0,255]?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

wqz960 commented 5 years ago

@FunkyKoki another question: for the regressor, the input of the network is the concatenate of raw image and boundary heat maps whose size is (batch, 16,256,2556) or others, I do not what are the feature maps. Thank you!!!

FunkyKoki commented 5 years ago

cannot understand your question very well, but if I did not forget, the regressor input is acquired by "input image fusion" as described in the paper

---Original--- From: "wqz960"notifications@github.com Date: Sat, Jun 8, 2019 17:38 PM To: "FunkyKoki/Look_At_Boundary_PyTorch"Look_At_Boundary_PyTorch@noreply.github.com; Cc: "Mention"mention@noreply.github.com;"Champagne_Jin"643683905@qq.com; Subject: Re: [FunkyKoki/Look_At_Boundary_PyTorch] Hi！ how to draw boundary heat map? (#6)

@FunkyKoki another question: for the regressor, the input of the network is the concatenate of raw image and boundary heat maps whose size is (batch, 16,256,2556) or others, I do not what are the feature maps. Thank you!!!

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

wqz960 commented 5 years ago

Hi @FunkyKoki the input fusion image for regressor seems different with the original paper, H = I ⊕ (M1 ⊗ I) ⊕ ... ⊕ (MT ⊗ I) which is ((13+1)*3, 256, 256) size, but your network is only support gray picture, which is (14,256,256). Am I mistaken in this respect?

FunkyKoki commented 5 years ago

indeed, the author did not follow the formula he gave. you should check the prototxt file he gave, and that is what i have done.

---Original--- From: "wqz960"notifications@github.com Date: Sun, Jun 9, 2019 15:14 PM To: "FunkyKoki/Look_At_Boundary_PyTorch"Look_At_Boundary_PyTorch@noreply.github.com; Cc: "Mention"mention@noreply.github.com;"Champagne_Jin"643683905@qq.com; Subject: Re: [FunkyKoki/Look_At_Boundary_PyTorch] Hi！ how to draw boundary heat map? (#6)

Hi @FunkyKoki the input fusion image for regressor seems different with the original paper, H = I ⊕ (M1 ⊗ I) ⊕ ... ⊕ (MT ⊗ I) which is ((13+1)*3, 256, 256) size, but your network is only support gray picture, which is (14,256,256). Am I mistaken in this respect?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.