search

Xharlie / DISN

(latest updates and bug fixed) DISN: Deep Implicit Surface Network for High-quality Single-view 3D Reconstruction
183 stars 27 forks source link

some questions about code and training #12

Open asurada404 opened 4 years ago

asurada404 commented 4 years ago

Q1: TOTAL_POINTS / 214669.0 in test/create_sdf.py(line 72) Why did you divide the TOTAL_POINTS by 214669.0? Is it because of the amount of GPU memory? How did you get the number 214669.0？

Q2: How many epochs were trained when training the camera estimation and SDF estimation with ground true camera pose respectively? Q3: How long does each epoch take when you training SDF estimation with ground true camera pose？ Thanks a lot^_^

Xharlie commented 4 years ago

A1: it is about the memory, feel free to change. A2,A3: it takes at least 80 epochs, i can't recall precisely.

AlexsaseXie commented 3 years ago

As you mentioned in #https://github.com/Xharlie/DISN/issues/3, it should be taking no more than 3 days to train. While I used a single 2080ti and trained for 3 days. I found the sdf loss stop dropping. But the final trained epoch count is only 23. Do you mean that I need to train more epochs to converge?