RuntimeError: Sizes of tensors must match except in dimension 2. Got 316 and 313 (The offending index is 0

[kihoon71]

Hi. I'm trying to use custom dataset with your model.
I've changed few codes where I think I have to, like
point_cloud_range, classes, etc.

When I run the command

---

python train.py --data_root own_data

I get error message below.

---

RuntimeError: Sizes of tensors must match except in dimension 2. Got 316 and 313 (The offending index is 0)

can I get any help?

[zhulf0804]

Hello, @kihoon71, which line causes the error ?

[kihoon71]

Instead of my team mate, i did upload this issue. He must know better about this issue than me, so i will mention him here.

@fb059

[fb059]

Hi, I'm the one who asked for it.

The error occures in line 181 of pointpillars.py.

I'll put total error message below.

Traceback (most recent call last):
File "train.py", line 222, in
main(args)
File "train.py", line 89, in main
pointpillars(batched_pts=batched_pts,
File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "/workspace/PointPillars/model/pointpillars.py", line 393, in forward
x = self.neck(xs)
File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "/workspace/PointPillars/model/pointpillars.py", line 181, in forward
out = torch.cat(outs, dim=1)
RuntimeError: Sizes of tensors must match except in dimension 2. Got 316 and 313 (The offending index is 0)

[zhulf0804]

Hi, @fb059 , could you help to provide your config for voxel_size and point_cloud_range ?
And then print(xi.size()) in

PointPillars/model/pointpillars.py

Line 179 in 57af3f5

xi = self.decoder_blocks[i](x[i]) # (bs, 128, 248, 216)

[fb059]

voxel_size is
[0.25, 0.25, 8]

and point_cloud_range is
[-50, -50, -5, 50, 50, 3].

And when I print(xi.size()) in pointpillars.py, the output is
torch.Size([6, 128, 200, 200])

[zhulf0804]

voxel_size is [0.25, 0.25, 8]

and point_cloud_range is [-50, -50, -5, 50, 50, 3].

And when I print(xi.size()) in pointpillars.py, the output is torch.Size([6, 128, 200, 200])

Hi, so how 316 and 313 are generated, when your x_i.size() is [6, 128, 200, 200]?

[fb059]

Sorry, my mistake.
I changed configs in pointpillars.py to

            nclasses=2, 
             voxel_size=[0.25, 0.25, 8],
             point_cloud_range=[-50, -50, -5, 50, 50, 3],
             max_num_points=64,
             max_voxels=(30000, 40000)):

and after that change

RuntimeError: Sizes of tensors must match except in dimension 2. Got 316 and 313 (The offending index is 0)

no longer occurs.
But now I get different error described below.

---

Traceback (most recent call last):
File "train.py", line 215, in
main(args)
File "train.py", line 154, in main
pointpillars(batched_pts=batched_pts,
File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "/workspace/PointPillars/model/pointpillars.py", line 396, in forward
pillar_features = self.pillar_encoder(pillars, coors_batch, npoints_per_pillar)
File "/opt/conda/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
result = self.forward(*input, **kwargs)
File "/workspace/PointPillars/model/pointpillars.py", line 92, in forward
pooling_features = torch.max(features, dim=-1)[0] # (p1 + p2 + ... + pb, out_channels)
RuntimeError: cannot perform reduction function max on tensor with no elements because the operation does not have an identity

[zhulf0804]

Hi @fb059, it seems there are no elements in tensor features.
Please help check features.size() in

PointPillars/model/pointpillars.py

Line 76 in 57af3f5

features = torch.cat([pillars, offset_pt_center, x_offset_pi_center, y_offset_pi_center], dim=-1) # (p1 + p2 + ... + pb, num_points, 9)

and

PointPillars/model/pointpillars.py

Line 91 in 57af3f5

features = F.relu(self.bn(self.conv(features))) # (p1 + p2 + ... + pb, out_channels, num_points)

[fb059]

Hi, @zhulf0804
First, I misunderstood my point_cloud_range(min x,y,z values and max x,y,z values), so I fixed it right.

point_cloud_range=[-39.9442, -39.7883, -2.3986, 39.8011, 39.8065, 6.9456],

And as you recommanded, I checked features.size()
The first features.size() looks like

torch.Size([583, 64, 9])
torch.Size([892, 64, 9])
torch.Size([879, 64, 9])
torch.Size([905, 64, 9])
torch.Size([626, 64, 9])
torch.Size([789, 64, 9])
torch.Size([742, 64, 9])
torch.Size([697, 64, 9])
torch.Size([632, 64, 9])
torch.Size([636, 64, 9])
torch.Size([623, 64, 9])
torch.Size([644, 64, 9])
torch.Size([679, 64, 9])
torch.Size([916, 64, 9])
torch.Size([871, 64, 9])
torch.Size([720, 64, 9])
torch.Size([704, 64, 9])
torch.Size([896, 64, 9])
torch.Size([833, 64, 9])
torch.Size([855, 64, 9])
torch.Size([880, 64, 9])
torch.Size([762, 64, 9])
torch.Size([872, 64, 9])
torch.Size([642, 64, 9])
torch.Size([825, 64, 9])
torch.Size([634, 64, 9])
torch.Size([751, 64, 9])
torch.Size([608, 64, 9])
torch.Size([474, 64, 9])
torch.Size([663, 64, 9])
torch.Size([540, 64, 9])
torch.Size([982, 64, 9])
torch.Size([772, 64, 9])
torch.Size([813, 64, 9])
torch.Size([0, 64, 9])
Traceback (most recent call last):

And the second features.size() looks like

torch.Size([583, 64, 9])
torch.Size([583, 64, 64])
torch.Size([892, 64, 9])
torch.Size([892, 64, 64])
torch.Size([879, 64, 9])
torch.Size([879, 64, 64])
torch.Size([905, 64, 9])
torch.Size([905, 64, 64])
torch.Size([626, 64, 9])
torch.Size([626, 64, 64])
torch.Size([789, 64, 9])
torch.Size([789, 64, 64])
torch.Size([742, 64, 9])
torch.Size([742, 64, 64])
torch.Size([697, 64, 9])
torch.Size([697, 64, 64])
torch.Size([632, 64, 9])
torch.Size([632, 64, 64])
torch.Size([636, 64, 9])
torch.Size([636, 64, 64])
torch.Size([623, 64, 9])
torch.Size([623, 64, 64])
torch.Size([644, 64, 9])
torch.Size([644, 64, 64])
torch.Size([679, 64, 9])
torch.Size([679, 64, 64])
torch.Size([916, 64, 9])
torch.Size([916, 64, 64])
torch.Size([871, 64, 9])
torch.Size([871, 64, 64])
torch.Size([720, 64, 9])
torch.Size([720, 64, 64])
torch.Size([704, 64, 9])
torch.Size([704, 64, 64])
torch.Size([896, 64, 9])
torch.Size([896, 64, 64])
torch.Size([833, 64, 9])
torch.Size([833, 64, 64])
torch.Size([787, 64, 9])
torch.Size([787, 64, 64])
torch.Size([855, 64, 9])
torch.Size([855, 64, 64])
torch.Size([880, 64, 9])
torch.Size([880, 64, 64])
torch.Size([762, 64, 9])
torch.Size([762, 64, 64])
torch.Size([872, 64, 9])
torch.Size([872, 64, 64])
torch.Size([642, 64, 9])
torch.Size([642, 64, 64])
torch.Size([825, 64, 9])
torch.Size([825, 64, 64])
torch.Size([634, 64, 9])
torch.Size([634, 64, 64])
torch.Size([751, 64, 9])
torch.Size([751, 64, 64])
torch.Size([608, 64, 9])
torch.Size([608, 64, 64])
torch.Size([474, 64, 9])
torch.Size([474, 64, 64])
torch.Size([663, 64, 9])
torch.Size([663, 64, 64])
torch.Size([935, 64, 9])
torch.Size([935, 64, 64])
torch.Size([540, 64, 9])
torch.Size([540, 64, 64])
torch.Size([982, 64, 9])
torch.Size([982, 64, 64])
torch.Size([772, 64, 9])
torch.Size([813, 64, 9])
torch.Size([813, 64, 64])
torch.Size([0, 64, 9])
torch.Size([0, 64, 64])
Traceback (most recent call last):

[zhulf0804]

Hi @fb059, as the final feature'size is [0, 64, 9], it seems that the data is abnormal. Please help check it. Thanks.