Please init pose by register first

[shzcuber]

Hello!

I installed FoundationPose with Ubuntu 20.04 and docker. I'm trying to run the run_demo.py script. Initially, I ran into the same memory issue as #26, so I changed my shorter side value to 10 (I can't do anything much above that value). Now, I'm getting this issue:

Any ideas on what the issue is?

Here's the error:

[__init__()] self.h5_file:
[__init__()] Using pretrained model from /home/yori/Documents/FoundationPose/learning/training/../../weights/2023-10-28-18-33-37/model_best.pth
[__init__()] init done
[reset_object()] self.diameter:0.19646325799497472, vox_size:0.009823162899748735
[reset_object()] self.pts:torch.Size([607, 3])
[reset_object()] reset done
[make_rotation_grid()] cam_in_obs:(42, 4, 4)
[make_rotation_grid()] rot_grid:(252, 4, 4)
num original candidates = 252
num of pose after clustering: 252
[make_rotation_grid()] after cluster, rot_grid:(252, 4, 4)
[make_rotation_grid()] self.rot_grid: torch.Size([252, 4, 4])
[<module>()] estimator initialization done
[<module>()] i:0
[register()] Welcome
Module Utils load on device 'cuda:0' took 3.69 ms
[register()] valid too small, return
[<module>()] i:1
[track_one()] Please init pose by register first
Traceback (most recent call last):
  File "run_demo.py", line 63, in <module>
    pose = est.track_one(rgb=color, depth=depth, K=reader.K, iteration=args.track_refine_iter)
  File "/home/yori/Documents/FoundationPose/estimater.py", line 253, in track_one
    raise RuntimeError
RuntimeError

Please let me know if there's any more information that could be provided.

[shzcuber]

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 198.00 MiB (GPU 0; 3.81 GiB total capacity; 1.99 GiB already allocated; 196.25 MiB free; 2.25 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Here's my OutOfMemoryError if that helps.

[GavinYang5]

It looks like the point cloud obtained from the mask you entered is too sparse and the registration is not successful.
Please check #44

[shzcuber]

Sorry, do you mind elaborating on what you mean by the mask I entered?

Not sure if this helps, but all the data I'm using is the demo_data as mentioned in the README.md - https://drive.google.com/drive/folders/1pRyFmxYXmAnpku7nGRioZaKrVJtIsroP?usp=sharing.

[GavinYang5]

Because you change the shorter edge value to 10, this will cause your depth image to be scaled out too small, resulting in no depth data. Suggest trying FoundationPose on a GPU with larger memory.