fix(config): range divide voxel_size need to be a INT for initial tensor size.

Author: Kin-Zhang

1_train.py

@@ -28,10 +28,19 @@
 from scripts.network.dataloader import HDF5Dataset, collate_fn_pad
 from scripts.pl_model import ModelWrapper
 
-@hydra.main(version_base=None, config_path="conf", config_name="config")
-def main(cfg):
+def precheck_cfg_valid(cfg):
     if cfg.loss_fn == 'seflowLoss' and cfg.add_seloss is None:
         raise ValueError("Please specify the self-supervised loss items for seflowLoss.")
+    if (cfg.point_cloud_range[3] - cfg.point_cloud_range[0]) % cfg.voxel_size[0] != 0 or \
+       (cfg.point_cloud_range[4] - cfg.point_cloud_range[1]) % cfg.voxel_size[1] != 0 or \
+       (cfg.point_cloud_range[5] - cfg.point_cloud_range[2]) % cfg.voxel_size[2] != 0:
+        # For example: 51.2/0.2=256 good, 51.2/0.3=170.67 wrong.
+        raise ValueError("The voxel size should be able to divide the point cloud range to a INT.")
+    return cfg
+
+@hydra.main(version_base=None, config_path="conf", config_name="config")
+def main(cfg):
+    precheck_cfg_valid(cfg)
     pl.seed_everything(cfg.seed, workers=True)
 
     train_dataset = HDF5Dataset(cfg.train_data, n_frames=cfg.num_frames, dufo=(cfg.loss_fn == 'seflowLoss'))

dataprocess/README.md

@@ -121,12 +121,6 @@ All these preprocess scripts will generate the same format `.h5` file. The file
 File: `[*:logid].h5` file named in logid. Every timestamp is the key of group (f[key]).
 
 ```python
-def process_log(data_dir: Path, mode: str, scene_num_id: int, output_dir: Path, n: Optional[int] = None) :
-    def create_group_data(group, pc, pose, gm = None, flow_0to1=None, flow_valid=None, flow_category=None, ego_motion=None):
-        group.create_dataset('lidar', data=pc.astype(np.float32))
-        group.create_dataset('pose', data=pose.astype(np.float64))
-
-
 def process_log(data_dir: Path, log_id: str, output_dir: Path, n: Optional[int] = None) :
     def create_group_data(group, pc, gm, pose, flow_0to1=None, flow_valid=None, flow_category=None, ego_motion=None):
         group.create_dataset('lidar', data=pc.astype(np.float32))
@@ -138,7 +132,6 @@ def process_log(data_dir: Path, log_id: str, output_dir: Path, n: Optional[int]
             group.create_dataset('flow_is_valid', data=flow_valid.astype(bool))
             group.create_dataset('flow_category_indices', data=flow_category.astype(np.uint8))
             group.create_dataset('ego_motion', data=ego_motion.astype(np.float32))
-
 ```
 
 After preprocessing, all data can use the same dataloader to load the data. As already in our DeFlow code.