Skip to content

feat(BEVFusion): added a positional encoding-like feature extractor #16

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 3 commits into
base: main
Choose a base branch
from
Open

feat(BEVFusion): added a positional encoding-like feature extractor #16

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
7388dd8
feat: added a positional encoding-like feature extractor before the f…
knzo25 Mar 13, 2025
73d1e52
chore: removed prints
knzo25 Mar 13, 2025
71cbdfa
Merge branch 'main' into feat/bevfusion_pos_enc
Shin-kyoto May 14, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
29 changes: 27 additions & 2 deletions projects/BEVFusion/bevfusion/sparse_encoder.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,9 @@
else:
from mmcv.ops import SparseConvTensor

import numpy as np
import torch


@MODELS.register_module()
class BEVFusionSparseEncoder(SparseEncoder):
Expand Down Expand Up @@ -44,6 +47,10 @@ class BEVFusionSparseEncoder(SparseEncoder):
def __init__(
self,
in_channels,
aug_features,
aug_features_min_values,
aug_features_max_values,
num_aug_features,
sparse_shape,
order=("conv", "norm", "act"),
norm_cfg=dict(type="BN1d", eps=1e-3, momentum=0.01),
Expand All @@ -58,6 +65,10 @@ def __init__(
assert block_type in ["conv_module", "basicblock"]
self.sparse_shape = sparse_shape
self.in_channels = in_channels
self.aug_features = aug_features
self.aug_features_min_values = torch.tensor(aug_features_min_values).cuda()
self.aug_features_max_values = torch.tensor(aug_features_max_values).cuda()
self.num_aug_features = num_aug_features
self.order = order
self.base_channels = base_channels
self.output_channels = output_channels
Expand All @@ -68,12 +79,16 @@ def __init__(
self.return_middle_feats = return_middle_feats
# Spconv init all weight on its own

if aug_features:
self.in_channels = in_channels * num_aug_features * 2
self.exponents = 2 ** torch.arange(0, num_aug_features).to(torch.device("cuda")).float()

assert isinstance(order, tuple) and len(order) == 3
assert set(order) == {"conv", "norm", "act"}

if self.order[0] != "conv": # pre activate
self.conv_input = make_sparse_convmodule(
in_channels,
self.in_channels,
self.base_channels,
3,
norm_cfg=norm_cfg,
Expand All @@ -84,7 +99,7 @@ def __init__(
)
else: # post activate
self.conv_input = make_sparse_convmodule(
in_channels,
self.in_channels,
self.base_channels,
3,
norm_cfg=norm_cfg,
Expand Down Expand Up @@ -127,6 +142,16 @@ def forward(self, voxel_features, coors, batch_size):
output features. When self.return_middle_feats is True, the
module returns middle features.
"""

if self.aug_features:
num_points = voxel_features.shape[0]
x = (voxel_features - self.aug_features_min_values.view(1, -1)) / (
self.aug_features_max_values - self.aug_features_min_values
).view(1, -1)
y = x.reshape(-1, 1) * np.pi * self.exponents.reshape(1, -1)
y = y.reshape(num_points, -1)
voxel_features = torch.cat([torch.cos(y), torch.sin(y)], dim=1)

coors = coors.int()
input_sp_tensor = SparseConvTensor(voxel_features, coors, self.sparse_shape, batch_size)
x = self.conv_input(input_sp_tensor)
Expand Down
Loading