PoseFree?????

[NNsauce]

hi, I saw the init() in Class PoseModel . I doubt that if there were some misunderstood in my mind about POSE FREE.
But IF NOT, where shoud I get that damn json file and npz file? where should I get gt_pose? flow? depth?

Sorry , I just feel too disappointed. If I am wrong, please correct me, thanks!

[wrld]

Our method is totally pose-free without the need of ground truth poses. The loaded pose is just for evaluation on the SCARED Dataset. The flow and depth are extracted from RGB images with RAFT model and Depth-Anything model. To quickly reproduce the results, you can download our demo example, thank you.

[NNsauce]

uh, good! I understand. Thanks so much, I apologize for my rudeness and impulsive.

[NNsauce]

My boss doesnt want me to keep going on mono RGB methods which I always keep a strong belief with.
Feels like, it would drag RAFT and DepthAnything to get flow and depth to get the correct data format, and then the training code has to be modified: remove pose eval and so on. This is just my guess...
your method is really a nice work, and I just want to reconstruct my 3D teeth model AS SOON AS POSSIBLE.
SO, do you mind releasing a simple tutorial about how to train custom data?