Cooperative perception allows connected vehicles and roadside infrastructure to share sensor observations, creating a fused scene representation beyond the capability of any single platform. However, most cooperative 3D object detectors use a uniform fusion strategy for all object classes, which limits their ability to handle the different geometric structures and point-sampling patterns of small and large objects. This problem is further reinforced by narrow evaluation protocols that often emphasize a single dominant class or only a few cooperation settings, leaving robust multi-class detection across diverse vehicle-to-everything interactions insufficiently explored. To address this gap, we propose a class-adaptive cooperative perception architecture for multi-class 3D object detection from LiDAR data. The model integrates four components: multi-scale window attention with learned scale routing for spatially adaptive feature extraction, a class-specific fusion module that separates small and large objects into attentive fusion pathways, bird's-eye-view enhancement through parallel dilated convolution and channel recalibration for richer contextual representation, and class-balanced objective weighting to reduce bias toward frequent categories. Experiments on the V2X-Real benchmark cover vehicle-centric, infrastructure-centric, vehicle-to-vehicle, infrastructure-to-infrastructure, and vehicle-to-infrastructure settings under identical backbone and training configurations. The proposed method consistently improves mean detection performance over strong intermediate-fusion baselines, with the largest gains on trucks, clear improvements on pedestrians, and competitive results on cars. These results show that aligning feature extraction and fusion with class-dependent geometry and point density leads to more balanced cooperative perception in realistic vehicle-to-everything deployments.
Figures & tables
Fig. 1: Overview of the proposed class-adaptive cooperative perception architecture for multi-class 3D object detection in V2X systems. The framework processes multi-agent LiDAR inputs through: (1) Multi-agent feature extraction via PointPillars backbone, (2) Multi-scale window attention with learned class-adaptive routing that adapts receptive fields to object scales (2×2, 4×4, 8×8, 16×16 windows), (3) Class-specific fusion block with dedicated pathways for small objects (pedestrians) and large objects (cars, trucks), (4) Multi-scale BEV enhancement using atrous spatial pyramid pooling (ASPP) with dilation rates [1, 3, 6, 12] and squeeze-and-excitation (SE) modules, and (5) 3D detection head with class-balanced loss optimization. The architecture addresses scale variation and class imbalance through differentiated feature routing and gradient reweighting.
Fig. 2: Bird’s-eye-view visualization of a representative V2X-Real scene showing multi-class 3D annotations: pedestrians (red, 25), cars (green, 16), and trucks (blue, 4) relative to ego vehicle (cyan). The dataset exhibits significant class imbalance and scale variations.
Fig. 3: Per-class detection performance comparison. Values are AP@50 (%) averaged across all five cooperation modes. Our method achieves the highest accuracy in every category, with the largest margin on truck detection (+5.8 pp on average over the strongest baseline per cooperation mode), confirming that class-differentiated fusion disproportionately benefits scale-diverse objects.
Fig. 4: Per-class performance profile in VC mode (AP@50). Each polygon represents one method. Our method (red, solid) achieves the largest enclosed area, reflecting the most balanced detection across all three classes. The pronounced Truck-axis expansion (+6.7 pp over the strongest baseline, AttFuse at 31.8%) highlights the benefit of scale-adaptive fusion for large objects.
Mode
Method
Vehicle
Pedestrian
Truck
mAP
AP@30
AP@50
Δ
AP@30
AP@50
Δ
AP@30
AP@50
Δ
AP@30
AP@50
VC
F-Cooper
62.9
49.5
-
26.8
9.9
-
35.0
23.6
-
41.6
27.7
AttFuse
58.5
55.1
-
29.4
12.9
-
37.0
31.8
-
41.6
33.3
V2X-ViT
67.3
61.5
-
29.8
12.1
-
38.6
29.0
-
45.3
34.2
Ours
67.8
62.8
+1.3
30.2
12.3
-0.6
44.6
38.5
+6.7
47.5
37.9
IC
F-Cooper
61.8
23.8
-
24.8
7.7
-
18.4
4.1
-
35.0
11.8
TABLE I: Comparative results on V2X-Real test set across all cooperation modes. All methods use PointPillars backbone with identical training procedures. Best results in bold , second best underlined . AP reported as percentage at IoU thresholds 0.3 and 0.5. Δ indicates the difference between our method and the best baseline for AP@50 (positive values in green , negative in red ).
Fig. 5: Qualitative comparison across three cooperation modes. Each row shows a representative scene: vehicle-centric (VC, top), vehicle-to-vehicle (V2V, middle), and infrastructure-centric (IC, bottom). Columns compare F-Cooper, AttFuse, V2X-ViT, and our method on identical frames. Green boxes denote ground truth annotations. Predicted bounding boxes are color-coded by class: orange (vehicle), red (pedestrian), magenta (truck). Our method produces more accurate localizations with fewer false positives and missed detections, particularly for underrepresented pedestrians and trucks. Best viewed in color with zoom.
Method
Params (M)
Memory (MB)
Inference (ms)
F-Cooper
8.07
30.8
13.1
AttFuse
8.07
30.8
15.6
V2X-ViT
13.47
51.4
289.9
Ours
15.89
60.6
345.8
TABLE II: Computational cost comparison. Params: total learnable parameters; Memory: model memory footprint; Inference: average per-sample inference time on NVIDIA A100.
Configuration
Vehicle
Pedestrian
Truck
mAP@50
Baseline (V2X-ViT)
52.3
11.1
25.3
29.5
+M1
56.8
13.6
39.4
36.6
+M1+M2
58.7
13.2
40.9
37.6
+M1+M2+M3 (Full)
56.4
12.9
38.0
35.8
Gain vs. Baseline
+4.1
+1.8
+12.7
+6.3
TABLE III: Ablation study on V2X-Real test set showing incremental contribution of each architectural component. Results reported for V2V cooperation mode at IoU threshold 0.5. M1: Multi-scale window attention + Class-specific fusion; M2: Multi-scale BEV enhancement; M3: Class-balanced loss.
Fig. 6: Qualitative ablation on a scene in 3D perspective view for v2v collaborative mode. (a) +M1: multi-scale attention and class-specific fusion; (b) +M2: multi-scale BEV enhancement; (c) +M3: full model with class-balanced loss. Green: ground truth; orange: vehicle; red: pedestrian; magenta: truck. Configurations (a) and (b) produce scattered pedestrian false positives (left region), while (c) yields cleaner, more calibrated predictions.
Fig. 7: Per-class AP@50 as a function of BEV distance, averaged across all cooperation modes. Vehicle detection (left) degrades with range for all methods, while our variants maintain higher accuracy throughout. Pedestrian detection (center) exhibits limited range sensitivity due to inherent LiDAR sparsity. Truck detection (right) shows consistent gains for our method at all distances, with the +M1+M2 configuration (Ours v2) achieving the best far-range performance owing to ASPP-based context aggregation.
Mode
Method
0–30 m (Near)
30–60 m (Mid)
60–100 m (Far)
Veh.
Ped.
Truck
mAP
Veh.
Ped.
Truck
mAP
Veh.
Ped.
Truck
mAP
VC
F-Cooper
59.9
9.9
16.7
28.8
47.2
8.8
31.8
29.3
39.8
12.3
42.7
31.6
AttFuse
81.6
14.6
35.7
44.0
51.2
12.6
29.5
31.1
27.0
11.8
43.9
27.5
V2X-ViT
81.0
13.7
24.9
39.8
60.6
12.2
31.7
34.8
36.5
10.8
39.9
29.0
Ours (v1)
83.0
15.1
49.5
49.2
62.9
13.1
42.8
39.6
36.2
12.5
53.5
34.0
Ours (v2)
85.6
15.2
44.0
48.3
60.1
13.0
46.0
39.7
41.2
11.1
59.9
37.4
TABLE IV: Range-stratified detection performance on V2X-Real test set. AP@50 (%) reported at three BEV distance ranges for all cooperation modes. Best in bold , second best underlined . Ours (v1): multi-scale attention + class-specific fusion; Ours (v2): +multi-scale BEV enhancement; Ours (Full): +class-balanced loss.
Zhejiang University, Hangzhou, China · Zhejiang Provincial Key Laboratory of Multi-Modal Communication Networks and Intelligent Information Processing, Hangzhou, China · China Jiliang University, Hangzhou, China