docs(README): update teflow, dogflow papers, wait for the merge.

hotfix(truckscene): bbx overlap on the same points, select the max one for assign flows.

Author: Kin-Zhang

README.md

@@ -11,10 +11,10 @@
 OpenSceneFlow is a codebase for point cloud scene flow estimation. 
 It is also an official implementation of the following papers (sorted by the time of publication):
 
-<!-- - **TeFlow: An Efficient Multi-frame Scene Flow Estimation Method**   
+- **TeFlow: Enabling Multi-frame Supervision for Self-Supervised Feed-forward Scene Flow Estimation**   
 *Qingwen Zhang, Chenhan Jiang, Xiaomeng Zhu, Yunqi Miao, Yushan Zhang, Olov Andersson, Patric Jensfelt*  
-Under Review   
-[ Strategy ] [ Self-Supervised ] - [ [OpenReview](https://openreview.net/forum?id=h70FLgnIAw) ] [ [Project](https://github.com/Kin-Zhang/TeFlow) ]&rarr; [here](#teflow) -->
+Conference on Computer Vision and Pattern Recognition (**CVPR**) 2026  
+[ Strategy ] [ Self-Supervised ] - [ [arXiv](https://arxiv.org/abs/2602.19053) ] [ [Project]() ]
 
 - **DeltaFlow: An Efficient Multi-frame Scene Flow Estimation Method**   
 *Qingwen Zhang, Xiaomeng Zhu, Yushan Zhang, Yixi Cai, Olov Andersson, Patric Jensfelt*  
@@ -26,6 +26,11 @@ Conference on Neural Information Processing Systems (**NeurIPS**) 2025 - Spotlig
 IEEE Transactions on Robotics (**T-RO**) 2025   
 [ Strategy ] [ Self-Supervised ] - [ [arXiv](https://arxiv.org/abs/2503.00803) ] [ [Project](https://kin-zhang.github.io/HiMo/) ] &rarr; [here](#seflow-1)
 
+- **DoGFlow: Self-Supervised LiDAR Scene Flow via Cross-Modal Doppler Guidance**   
+*Ajinkya Khoche, Qingwen Zhang, Yixi Cai, Sina Sharif Mansouri and Patric Jensfelt*    
+IEEE Robotics and Automation Letters (**RA-L**) 2026  
+[ Multi-Model ] [ Self-Supervised ] - [ [arXiv](https://arxiv.org/abs/2508.18506) ] [ [Project](https://ajinkyakhoche.github.io/DogFlow/) ]&rarr; [here](https://github.com/ajinkyakhoche/DoGFlow)
+
 - **VoteFlow: Enforcing Local Rigidity in Self-Supervised Scene Flow**   
 *Yancong Lin\*, Shiming Wang\*, Liangliang Nan, Julian Kooij, Holger Caesar*   
 Conference on Computer Vision and Pattern Recognition (**CVPR**) 2025  
@@ -46,7 +51,6 @@ International Conference on Robotics and Automation (**ICRA**) 2025
 European Conference on Computer Vision (**ECCV**) 2024  
 [ Strategy ] [ Self-Supervised ] - [ [arXiv](https://arxiv.org/abs/2407.01702) ] [ [Project](https://github.com/KTH-RPL/SeFlow) ] &rarr; [here](#seflow)
 
-
 - **DeFlow: Decoder of Scene Flow Network in Autonomous Driving**  
 *Qingwen Zhang, Yi Yang, Heng Fang, Ruoyu Geng, Patric Jensfelt*  
 International Conference on Robotics and Automation (**ICRA**) 2024  
@@ -62,6 +66,7 @@ Additionally, *OpenSceneFlow* integrates following excellent works: [ICLR'24 Zer
 - [x] [NSFP](https://arxiv.org/abs/2111.01253): NeurIPS 2021, faster 3x than original version because of [our CUDA speed up](assets/cuda/README.md), same (slightly better) performance.
 - [x] [FastNSF](https://arxiv.org/abs/2304.09121): ICCV 2023. SSL Optimization-based.
 - [x] [ICP-Flow](https://arxiv.org/abs/2402.17351): CVPR 2024. SSL Optimization-based.
+- [ ] [Floxels](https://arxiv.org/abs/2503.04718): CVPR 2025. SSL optimization-based. coding now but not yet ready for release as lower performance than reported. check [branch code](https://github.com/Kin-Zhang/OpenSceneFlow/tree/feature/floxels) for more details.
 - [ ] [EulerFlow](https://arxiv.org/abs/2410.02031): ICLR 2025. SSL optimization-based. In my plan, haven't coding yet.
 
 </details>
@@ -144,7 +149,7 @@ Train DeltaFlow with the leaderboard submit config. [Runtime: Around 18 hours in
 
 ```bash
 # total bz then it's 10x2 under above training setup.
-python train.py model=deltaFlow optimizer.lr=2e-3 epochs=20 batch_size=2 num_frames=5 loss_fn=deflowLoss train_aug=True "voxel_size=[0.15, 0.15, 0.15]" "point_cloud_range=[-38.4, -38.4, -3.2, 38.4, 38.4, 3.2]" +optimizer.scheduler.name=WarmupCosLR +optimizer.scheduler.max_lr=2e-3 +optimizer.scheduler.total_steps=20000
+python train.py model=deltaFlow optimizer.lr=2e-3 epochs=20 batch_size=2 num_frames=5 loss_fn=deflowLoss train_aug=True "voxel_size=[0.15, 0.15, 0.15]" "point_cloud_range=[-38.4, -38.4, -3, 38.4, 38.4, 3]" +optimizer.scheduler.name=WarmupCosLR +optimizer.scheduler.max_lr=2e-3 +optimizer.scheduler.total_steps=20000
 
 # Pretrained weight can be downloaded through (av2), check all other datasets in the same folder.
 wget https://huggingface.co/kin-zhang/OpenSceneFlow/resolve/main/deltaflow/deltaflow-av2.ckpt
@@ -211,7 +216,7 @@ python train.py model=deflow optimizer.lr=2e-4 epochs=9 batch_size=16 loss_fn=se
 wget https://huggingface.co/kin-zhang/OpenSceneFlow/resolve/main/seflow_best.ckpt
 ```
 
-#### VoteFLow
+#### VoteFlow
 Extra pakcges needed for VoteFlow, [pytorch3d](https://pytorch3d.org/) (prefer 0.7.7) and [torch-scatter](https://github.com/rusty1s/pytorch_scatter?tab=readme-ov-file) (prefer 2.1.2):
 
 ```bash
@@ -350,6 +355,13 @@ It is actively maintained and developed by the community (ref. below works).
 If you find it useful, please cite our works:
 
 ```bibtex
+@inproceedings{zhang2026teflow,
+  title = {{TeFlow}: Enabling Multi-frame Supervision for Self-Supervised Feed-forward Scene Flow Estimation},
+  author={Zhang, Qingwen and Jiang, Chenhan and Zhu, Xiaomeng and Miao, Yunqi and Zhang, Yushan and Andersson, Olov and Jensfelt, Patric},
+  year = {2026},
+  booktitle = {Proceedings of the IEEE/CVF conference on computer vision and pattern recognition},
+  pages = {},
+}
 @inproceedings{zhang2024seflow,
   author={Zhang, Qingwen and Yang, Yi and Li, Peizheng and Andersson, Olov and Jensfelt, Patric},
   title={{SeFlow}: A Self-Supervised Scene Flow Method in Autonomous Driving},
@@ -383,17 +395,21 @@ If you find it useful, please cite our works:
   year={2025},
   url={https://openreview.net/forum?id=T9qNDtvAJX}
 }
-@misc{zhang2025teflow,
-  title={{TeFlow}: Enabling Multi-frame Supervision for Feed-forward Scene Flow Estimation},
-  author={Zhang, Qingwen and Jiang, Chenhan and Zhu, Xiaomeng and Miao, Yunqi and Zhang, Yushan and Andersson, Olov and Jensfelt, Patric},
-  year={2025},
-  url={https://openreview.net/forum?id=h70FLgnIAw}
-}
 ```
 
 And our excellent collaborators works contributed to this codebase also:
 
 ```bibtex
+@article{khoche2026dogflow,
+  author={Khoche, Ajinkya and Zhang, Qingwen and Cai, Yixi and Mansouri, Sina Sharif and Jensfelt, Patric},
+  journal = {IEEE Robotics and Automation Letters},
+  title = {{DoGFlow}: Self-Supervised LiDAR Scene Flow via Cross-Modal Doppler Guidance},
+  year = {2026},
+  volume = {11},
+  number = {3},
+  pages = {3836-3843},
+  doi = {10.1109/LRA.2026.3662592},
+}
 @article{kim2025flow4d,
   author={Kim, Jaeyeul and Woo, Jungwan and Shin, Ukcheol and Oh, Jean and Im, Sunghoon},
   journal={IEEE Robotics and Automation Letters}, 

dataprocess/extract_truckscenes.py

@@ -68,9 +68,12 @@ def create_group_data(group, pc, pose, lidar_id, lidar_center, gm = None, flow_0
     def compute_flow_simple(data_fn, pc0, pose0, pose1, ts0, ts1, sample_ann_list, dclass, DataNameMap=ManNamMap):
         # compute delta transform between pose0 and pose1
         ego1_SE3_ego0 = npcal_pose0to1(pose0, pose1)
-        # flow due to ego motion
-        flow = np.zeros_like(pc0[:,:3])
-        flow = pc0[:,:3] @ ego1_SE3_ego0[:3,:3].T + ego1_SE3_ego0[:3,3] - pc0[:,:3] # pose flow
+        # flow due to ego motion (baseline for all points)
+        ego_flow = pc0[:,:3] @ ego1_SE3_ego0[:3,:3].T + ego1_SE3_ego0[:3,3] - pc0[:,:3]
+        
+        # object flow (without ego motion), used to track max flow magnitude
+        obj_flow_all = np.zeros_like(pc0[:,:3])
+        obj_flow_magnitude = np.zeros(len(pc0), dtype=np.float32)
 
         valid = np.ones(len(pc0), dtype=np.bool_)
         classes = np.zeros(len(pc0), dtype=np.uint8)
@@ -96,13 +99,21 @@ def compute_flow_simple(data_fn, pc0, pose0, pose1, ts0, ts1, sample_ann_list, d
             classes[points_in_box_mask] = CATEGORY_TO_INDEX[DataNameMap[cls]]
 
             if np.sum(points_in_box_mask) > 5:
-                obj_flow = np.ones_like(pc0[points_in_box_mask,:3]) * ann_vel * delta_t
-                flow[points_in_box_mask] += obj_flow
-                instances[points_in_box_mask] = (dclass[id_]+1)
+                obj_flow = ann_vel * delta_t
+                obj_flow_mag = np.linalg.norm(obj_flow)
+                
+                # For overlapping boxes, keep the flow with higher magnitude
+                higher_flow_mask = points_in_box_mask & (obj_flow_mag > obj_flow_magnitude)
+                obj_flow_all[higher_flow_mask] = obj_flow
+                obj_flow_magnitude[higher_flow_mask] = obj_flow_mag
+                instances[higher_flow_mask] = (dclass[id_]+1)
                 id_ += 1
             else:
                 valid[points_in_box_mask] = False
 
+        # Final flow = ego motion + object flow (with max magnitude selection)
+        flow = ego_flow + obj_flow_all
+
         return {'flow_0_1': flow, 'valid_0': valid, 'classes_0': classes, 
                 'ego_motion': ego1_SE3_ego0, 'flow_instance_id': instances}
 
@@ -183,16 +194,18 @@ def compute_flow_simple(data_fn, pc0, pose0, pose1, ts0, ts1, sample_ann_list, d
             # lidar_dt = lidar_dt[not_close]
             is_ground_0 = np.array(mygroundseg.run(points[:, :3]))
 
+            # HARDCODE: for TruckScenes, add all points below 0.2m as ground
+            is_ground_0 = is_ground_0 | (points[:,2] < 0.2)
+
+            group = f.create_group(str(ts0))
             if cnt == len(full_sweep_data_dict[SelectedSensor[-1]]) - 1:
-                group = f.create_group(str(ts0))
                 create_group_data(group=group, pc=points, gm=is_ground_0.astype(np.bool_), pose=pose0, \
                                 lidar_id=lidar_id, lidar_center=lidar_center)
             else:
                 sweep_data_next = full_sweep_data_dict[SelectedSensor[-1]][cnt+1]
                 ts1 = sweep_data_next['timestamp']
                 pose1 = get_pose(mants, sweep_data_next, w2stf=False)
 
-                group = f.create_group(str(ts0))
                 # annotated frame, compute flow
                 if sweep_data['is_key_frame'] and sweep_data['prev'] != "":
                     curr_scene_ann = mants.get_boxes(sweep_data['token'])
@@ -242,7 +255,7 @@ def process_logs(data_mode, data_dir: Path, scene_list: list, output_dir: Path,
             res = list(tqdm(p.imap_unordered(proc, args), total=len(scene_list), ncols=100))
 
 def main(
-    data_dir: str = "/home/kin/data/truckscenes/man-truckscenes",
+    data_dir: str = "/home/kin/data/man-demo/man-truckscenes",
     mode: str = "v1.0-mini",
     output_dir: str ="/home/kin/data/truckscenes/h5py",
     nproc: int = (multiprocessing.cpu_count() - 1),