Repository navigation
Feat/pandaset support - #3147
Open
SheheryarRamzan wants to merge 4 commits into
Open
SheheryarRamzan wants to merge 4 commits into
SheheryarRamzan wants to merge 4 commits into
Conversation
added 3 commits
August 25, 2026 11:07
Add PandaSet dataset integration for cross-dataset evaluation: - PandaSetDataset class with nuScenes 10-class label mapping - On-the-fly point cloud loading with world-to-ego transform - nuScenes-style evaluation metric (center-distance AP, NDS) - Dataset converter for dual-LiDAR split (Pandar64/PandarGT) - PointPillars evaluation configs for both sensors - Pre-conversion tool for .bin/.pcd formats
- Add configs/_base_/datasets/pandaset-3d.py base dataset config - Add docs/en/advanced_guides/datasets/pandaset.md with full guide - Fix data_root to use standard data/pandaset/ convention - Rewrite convert_pandaset_to_bin_pcd.py with argparse (no hardcoded paths) - Fix docstring example in pandaset_converter.py
|
Sheheryar Ramzan seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account. You have signed the CLA already but the status is still pending? Let us recheck it. |
Author
|
minimum_version_cpu fails while installing mmengine@main on Python 3.7.4. Current MMEngine setup.py uses importlib.metadata, which is 3.8+. This is unrelated to the PandaSet changes. Please pin MMEngine or bump the minimum CI Python. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.
Motivation
MMDetection3D currently lacks support for the PandaSet dataset, which is a valuable autonomous driving dataset featuring a dual-LiDAR system (Hesai Pandar64 + PandarGT). PandaSet enables cross-dataset generalization testing — evaluating models trained on one dataset (e.g., nuScenes) against data from different LiDAR sensors. This is critical for developing robust LiDAR perception algorithms that generalize beyond their training distribution.
Modification
Adds full PandaSet dataset support for 3D object detection:
mmdet3d/datasets/pandaset_dataset.py— PandaSetDataset class with nuScenes 10-class label mappingmmdet3d/datasets/transforms/pandaset_loading.py— LoadPointsFromPandaSet transform (world-to-ego on-the-fly) + NormalizePointsIntensitymmdet3d/evaluation/metrics/pandaset_metric.py— nuScenes-style metric (center-distance AP, mATE, mASE, mAOE, NDS)tools/dataset_converters/pandaset_converter.py— Info pkl generator with dual-LiDAR splittools/convert_pandaset_to_bin_pcd.py— Optional .bin/.pcd pre-conversion toolconfigs/_base_/datasets/pandaset-3d.py— Base dataset configconfigs/pointpillars/*pandaset*.py— Evaluation configs for Pandar64 and PandarGTdocs/en/advanced_guides/datasets/pandaset.md— Full documentationCross-dataset results (nuScenes PointPillars FPN → PandaSet Pandar64):
BC-breaking (Optional)
No. This PR only adds new files and extends
__init__.pyimports. No existing code is modified in a breaking way.Use cases (Optional)
1. Cross-dataset generalization testing — Evaluate any nuScenes-trained 3D detector on PandaSet to measure sensor generalization:
python tools/test.py \ configs/pointpillars/pointpillars_hv_fpn_sbn-all_8xb4-2x_pandaset-eval.py \ checkpoints/hv_pointpillars_fpn_sbn-all_4x8_2x_nus-3d.pth2. Dual-LiDAR comparison — Compare model performance on 360 degree spinning LiDAR (Pandar64) vs forward-facing solid-state LiDAR (PandarGT)
3. Distance-based analysis — Built-in distance-bin evaluation (0-25m, 25-50m) shows near-range vs far-range performance automatically
Checklist