Knowledge & Dataset Distillation
Cross-Modal 3D Distillation Without Paired and Synchronized LiDAR Streams
Generated automatically from the limitations stated in 6 papers (CVPR, ECCV, NeurIPS), listed under Evidence. It is not a paper, and it does not come from papers submitted to CSPaper.
The problem
Existing cross-modal distillation frameworks for 3D perception strictly require synchronized, spatially calibrated LiDAR point clouds and pre-trained LiDAR teachers alongside camera or radar feeds during training. This strict precondition limits distillation to expensive research vehicles with high-end sensor rigs, preventing its application to the vast majority of production fleet datasets where LiDAR is absent, uncalibrated, or asynchronously logged. As a result, low-cost sensor suites cannot leverage pre-trained LiDAR representations unless rigid multimodal collection setups are maintained during training.
Why it matters
Enables training high-accuracy camera and radar 3D perception models using pre-trained LiDAR teachers without requiring simultaneous, calibrated LiDAR rigs on the training vehicles.
Ways to approach it
Prior-work checks are free with an account. Results someone already ran are shown to everyone.
- 1
Systematic sensitivity analysis: Benchmark representative LiDAR-to-camera distillation methods (e.g., BEV feature distillation) under controlled temporal desynchronization and spatial calibration offsets to quantify the precise point-pixel alignment tolerance before transfer collapses, measuring mAP drop on standard autonomous driving datasets.
- 2
Distributional cross-modal matching: Develop unpaired feature distillation that aligns camera BEV representations with LiDAR teacher representations via optimal transport or statistical domain alignment rather than exact point-pixel correspondence, measuring 3D detection mAP on unpaired train splits.
- 3
Asynchronous teacher transfer: Construct an offline geometric prior bank from pre-existing LiDAR datasets and distill into camera students via topological or structural similarity matching, evaluated on pure camera datasets lacking concurrent LiDAR.
Have a different approach?
Describe how you would tackle this problem and we'll look for papers that already do it. Free; your text stays private.
Why it might fail
Pure vision-based foundation models and monocular depth priors might advance rapidly enough to render teacher-student LiDAR transfer obsolete before unpaired distillation stabilizes.
Evidence
Each paper's own statement of the limitation, verbatim.
- Three Pillars Improving Vision Foundation Model Distillation for LidarCVPR 2024
Requires precise spatial and temporal calibration between lidar sensors and cameras to establish point-pixel correspondences.
- RadarDistill: Boosting Radar-based Object Detection Performance via Knowledge Distillation from LiDAR FeaturesCVPR 2024
Requires high-quality, synchronized LiDAR data during the training phase which may not be available in all radar datasets.
- LabelDistill: Label-guided Cross-modal Knowledge Distillation for Camera-based 3D Object DetectionECCV 2024
Requires a pretrained LiDAR detection head whose inverse can be approximated, plus LiDAR data at training time
- CRKD: Enhanced Camera-Radar Object Detection with Cross-modality Knowledge DistillationCVPR 2024
Requires high-quality LiDAR data during training, which may not be available for all datasets or vehicle configurations.
- STXD: Structural and Temporal Cross-Modal Distillation for Multi-View 3D Object DetectionNeurIPS 2023
Requires synchronized, spatially aligned LiDAR point clouds and multi-camera images during training, so it cannot be applied where LiDAR data is unavailable
- BEVDistill: Cross-Modal BEV Distillation for Multi-View 3D Object DetectionICLR 2023
Requires LiDAR point clouds and a pre-trained LiDAR teacher at training time, so it only applies where paired LiDAR-camera data exists
Nearest existing work
- DistillBEV: Boosting Multi-Camera 3D Object Detection with Cross-Modal Knowledge DistillationICCV 2023
- LabelDistill: Label-guided Cross-modal Knowledge Distillation for Camera-based 3D Object DetectionECCV 2024
- UniDistill: A Universal Cross-Modality Knowledge Distillation Framework for 3D Object Detection in Bird's-Eye ViewCVPR 2023
- Leveraging Vision-Centric Multi-Modal Expertise for 3D Object DetectionNeurIPS 2023
- CRKD: Enhanced Camera-Radar Object Detection with Cross-modality Knowledge DistillationCVPR 2024
- Three Pillars Improving Vision Foundation Model Distillation for LidarCVPR 2024
- RCTDistill: Cross-Modal Knowledge Distillation Framework for Radar-Camera 3D Object Detection with Temporal FusionICCV 2025
- VeXKD: The Versatile Integration of Cross-Modal Fusion and Knowledge Distillation for 3D PerceptionNeurIPS 2024
- BEVDistill: Cross-Modal BEV Distillation for Multi-View 3D Object DetectionICLR 2023
- FSD-BEV: Foreground Self-Distillation for Multi-view 3D Object DetectionECCV 2024
- MapDistill: Boosting Efficient Camera-based HD Map Construction via Camera-LiDAR Fusion Model DistillationECCV 2024
- STXD: Structural and Temporal Cross-Modal Distillation for Multi-View 3D Object DetectionNeurIPS 2023
- RadarDistill: Boosting Radar-based Object Detection Performance via Knowledge Distillation from LiDAR FeaturesCVPR 2024
- MonoTAKD: Teaching Assistant Knowledge Distillation for Monocular 3D Object DetectionCVPR 2025
- X3KD: Knowledge Distillation Across Modalities, Tasks and Stages for Multi-Camera 3D Object DetectionCVPR 2023
Related open problems
Knowledge & Dataset Distillation
Knowledge Distillation Under Strict Black-Box Teacher Preconditions
Current high-performance distillation techniques rely on white-box access to intermediate representations, full logit vectors, or direct control over the teacher's training dynamics. When the teacher is a proprietary API or closed-source system that only returns discrete text or top-1 predictions, these methods cannot run. Downstream practitioners are structurally blocked from transferring capabilities from frontier closed models into compact architectures using state-of-the-art distillation losses.
Knowledge & Dataset Distillation
Distillation Under Zero-Data and Unlabeled-Stream Preconditions
Current distillation frameworks depend on strict data preconditions, including access to original downstream training sets, in-domain unlabeled datasets, out-of-distribution reference data, or sample-level identity annotations. When proprietary, privacy, or security constraints prevent the release of both the underlying training set and external proxy data, these methods cannot execute at all. Furthermore, when teacher outputs arrive as an uncurated stream without identity labels, existing feature-bank and prototype-matching methods break down. Resolving these preconditions allows model compression and dataset synthesis in strictly zero-data or privacy-restricted deployments.
Knowledge & Dataset Distillation
Knowledge Distillation and Rollout Transfer from Black-Box Teachers Without Logit Access
High-performing knowledge distillation and policy transfer techniques rely on white-box teacher access to evaluate conditional entropy, top-$k$ predictive distributions, and log-probabilities during rollouts. This requirement structurally prevents distillation from state-of-the-art proprietary or API-only models that output only discrete decisions or text tokens without internal logits. Furthermore, existing workarounds often rely on fragile parametric assumptions, such as Gaussianity of teacher logits under data augmentation, which fail on complex, multi-modal output distributions.