Choose a model¶
Prepare ZOD, Waymo, or Iseauto, then open a model page below. Each page has dataset and input-mode tabs, a copyable complete JSON config, a downloadable JSON file, and the exact command to run it.
| Model | Example page | Input size |
|---|---|---|
| CLFT | Configs and commands | 384 × 384 |
| CLFTv2 | Configs and commands | 256 × 256 |
| DeepLabV3+ | Configs and commands | 256 × 256 |
| MaskFormer | Configs and commands | 256 × 256 |
| Mask2Former | Configs and commands | 256 × 256 |
Choose an input mode¶
| Mode | Input | ZOD annotation folder | Waymo / Iseauto folder |
|---|---|---|---|
| RGB | Camera image | annotation_camera_only |
annotation |
| LiDAR | Camera-aligned three-channel LiDAR projection | annotation_lidar_only |
annotation |
| Fusion | Both inputs | annotation_fusion |
annotation |
ZOD examples change the annotation set with the input mode. For a comparison
with identical targets across modes, set the same Dataset.annotation_path
in each config. fusion works for every model; the loader handles internal
mode names.
Find a config in the checkout¶
There are 45 combinations. Each file is complete and uses a built-in model preset; no parent config file is required. Dataset manifests provide class mappings and normalization: ZOD and Waymo train four classes, Iseauto three.
Next: Run and inspect the outputs or customize a run.