HavenFeng / ArtEq

Generalizing Neural Human Fitting to Unseen Poses With Articulated SE(3) Equivariance (ICCV2023)
https://arteq.is.tue.mpg.de/
Other
46 stars 3 forks source link

so3 sampling issues #1

Closed fefei69 closed 5 months ago

fefei69 commented 10 months ago

Hi, When I tried to run your training sample code, I got 2 issues. First, I can't find ArtEq/external/vgtk/vgtk/data/anchors/sphere12.ply as needed in ANCHOR_PATH = os.path.join(ROOT, 'data', 'anchors/sphere12.ply')
(in https://github.com/HavenFeng/ArtEq/blob/main/external/vgtk/vgtk/so3conv/functional.py )

So I just copy the sphere12.ply from other repo (https://github.com/dragonlong/equi-pose/blob/master/models/vgtk/data/anchors/sphere12.ply)

Second, After using that sphere12.ply file I got another error in https://github.com/HavenFeng/ArtEq/blob/main/external/vgtk/vgtk/functional/rotation.py

python src/train.py     --EPN_input_radius 0.4     --EPN_layer_num 2     --aug_type so3     --batch_size 2     --epochs 15     --gt_part_seg auto     --i 0     --kinematic_cond yes     --num_point 5000
adj_idx: [array([[4, 1, 5]]), array([[0, 2]]), array([[1, 3]]), array([[2, 4, 8]]), array([[0, 3, 9]]), array([[ 0, 10]]), array([[11, 12]]), array([[12, 13]]), array([[ 3, 14]]), array([[ 4, 10, 14]]), array([[ 9,  5, 15]]), array([[ 6, 16]]), array([[ 6,  7, 17]]), array([[ 7, 18]]), array([[ 9,  8, 19]]), array([[10, 19, 16]]), array([[11, 15, 17]]), array([[12, 16, 18]]), array([[13, 17, 19]]), array([[14, 15, 18]])]
Traceback (most recent call last):
File "/home/Desktop/ArtEq/external/vgtk/vgtk/functional/rotation.py", line 259, in icosahedron_so3_trimesh
R_adj = get_adjmatrix_trimesh(mesh, gsize)
File "/home/Desktop/ArtEq/external/vgtk/vgtk/functional/rotation.py", line 128, in get_adjmatrix_trimesh
face_adj =  np.vstack(adj_idx).astype(np.int32)
File "/home/anaconda3/envs/arteq/lib/python3.10/site-packages/numpy/core/shape_base.py", line 289, in vstack
return _nx.concatenate(arrs, 0, dtype=dtype, casting=casting)
ValueError: all the input array dimensions except for the concatenation axis must match exactly, but along dimension 1, the array at index 0 has size 3 and the array at index 1 has size 2

It seems like that the dimensions of some array in adj_idx are wrong so that they can't be stacked vertically but I'm not sure if it is the mesh file issue or code issue.

Thanks in advance!!

mathXin112 commented 10 months ago

The same issue,looking forward to a reply!

HavenFeng commented 10 months ago

Hi, thanks for pointing it out, we are working on it now and will release an update soon, stay tuned!

HavenFeng commented 5 months ago

Just updated!

mathXin112 commented 5 months ago

----------------The message has been received,best wishes!您好,邮件已经收到,祝好。