Hello, I want to ask a question about the network about a2fnet

[sollynoay]

Sorry for the late response. The code here is for generating the front view depth map. The 3D model is generated from the estimated depth map directly by coordinate transformation. Please feel free to ask if there are any questions.

[bbbmbbbm]

thank you for your reply!
Do you use the traditional method of feature extraction and feature matching to generate 3D models through coordinate transformation? Or is there a special software program?
Thank you again for your help, and I wish you progress in your studies, smooth work and successful research.

[sollynoay]

Thank you for your question. The estimated depth map from the network is similar to the depth map of an optical camera. In our case, the width direction refers to the azimuth angle and the height direction refers to the elevation angle and each value of the pixel indicates the range. Each pixel can be then transformed to a 3D point from polar coordinate to euclidean coordinate transformation. And finally, a point cloud can be generated.

[bbbmbbbm]

Thank you for your reply, I am prepared to study, I may ask you if I don't understand.

[bbbmbbbm]

Hello, I would like to ask you a question about the dataset. What is the sfront.txt file in your dataset for? Is it the ground truth of the depth value?
Is front_resize.jpg a synthetic depth image? What is his role in the dataset?

Looking forward to your answer, thank you very much

[sollynoay]

Hello, sfront.txt and front_resize.jpg have the exact same information. The difference is that an 8-bit .jpg file is not very precise and information may be lost during compression. The data in front.txt is more precise, actually, we use sfront.txt to train the A2FNet.

[bbbmbbbm]

Thank you for your reply, I learned a lot
Sorry I have another question, may I ask you, how did you generate the sfront.txt file in your dataset? Is it generated with cycleGAN network?
Looking forward to your reply, thanks again