v2.0.0 API Rewrite (#4)

* Bunch of stuff is still broken, but AV2 Demo 3D works for RGB + Lidar

* Fully passing tests

* Changed Item convention to Frame

* Fixed backwards frame transform issue with AV2 RGB

* Added reshape support to crop AV2's traffic cam light

* Added valid pixels mask

* Added proper tests for image crop

* Added pulls for AV2 Small to do prep for full sequence tests

* Added separate setup step

* Cleanup names in testing infra

* Added tests for loading full sequences

* Added Causal and NonCausal loader

* Stopped deleting ground points entirely

* Added proper downscaling and transpose support to AV2 dataloader

* Bump to v2.0.0

* Cleaned up demo and eval code

* Cleanup readme

* Added floor and ceil to crop resize

.github/workflows/python_install.yml

@@ -1,9 +1,9 @@
-name: Python Package Install
+name: Bucketed Scene Flow Eval full tests
 
 on: [push]
 
 jobs:
-  install:
+  run-test:
     runs-on: ubuntu-latest
 
     steps:
@@ -18,11 +18,15 @@ jobs:
       with:
         python-version: '3.10'
 
-    - name: Install Dependencies
+    - name: Pip Install
       run: |
         python -m pip install --upgrade pip
         pip install -e .
 
+    - name: Download and Setup Data Files
+      run: |
+        ./tests/setup.sh
+
     - name: Run Tests
       run: |
         ./tests/integration_tests.sh

README.md

@@ -10,7 +10,7 @@ This repo provides:
 Currently supported datasets:
 
  - Argoverse 2 (Human Labeled and [NSFP Pseudolabeled](https://github.com/kylevedder/BucketedSceneFlowEval/blob/master/docs/GETTING_STARTED.md#argoverse-2-nsfp-pseudolabels-new))
- - Waymo Open
+ - Waymo Open (LiDAR only)
 
 
 If you use this repository as part of a publication, please cite:
@@ -45,23 +45,23 @@ We provide a demo script which shows off the various features of the API.
 To render the lidar and multiple camera views of an Argoverse 2 sequence in 3D, run:
 
 ```
-python scripts/demo.py --dataset Argoverse2SceneFlow --root_dir /efs/argoverse2/val/
+python scripts/demo_3d.py --dataset Argoverse2CausalSceneFlow --root_dir /efs/argoverse2/val/ --with_rgb --sequence_length 4
 ```
 
 ![Argoverse 2 MultiCam](docs/imgs/av2_multicam.png)
 
 To render RGB frames with lidar imposed on top, run:
 
 ```
-python scripts/demo_rgb.py --dataset Argoverse2SceneFlow --mode project_lidar --save_dir /efs/av2_camera_render/ --reduction_factor 16 --root_dir /efs/argoverse2/val
+python scripts/demo_rgb.py --dataset Argoverse2SceneFlow --mode project_lidar  --reduction_factor 16 --root_dir /efs/argoverse2/val --sequence_length 150 --save_dir /efs/av2_camera_render/
 ```
 
 ![Argoverse 2 LiDAR](docs/imgs/av2_lidar.gif)
 
 To render the flow field of an Argoverse 2 sequence, run:
 
 ```
-python scripts/demo_rgb.py --dataset Argoverse2SceneFlow --mode project_flow --save_dir /efs/av2_camera_render/ --reduction_factor 16 --root_dir /efs/argoverse2/val --flow_dir <path to method flow output>
+python scripts/demo_rgb.py --dataset Argoverse2SceneFlow --mode project_flow  --reduction_factor 16 --root_dir /efs/argoverse2/val --sequence_length 150 --save_dir /efs/av2_camera_render/ --flow_dir <path to method flow output>
 ```
 
 ![Argoverse 2 Flow](docs/imgs/av2_gt_flow.gif)

bucketed_scene_flow_eval/datasets/__init__.py

@@ -1,11 +1,23 @@
-from bucketed_scene_flow_eval.datasets.argoverse2 import Argoverse2SceneFlow
-from bucketed_scene_flow_eval.datasets.waymoopen import WaymoOpenSceneFlow
+from bucketed_scene_flow_eval.datasets.argoverse2 import (
+    Argoverse2CausalSceneFlow,
+    Argoverse2NonCausalSceneFlow,
+)
+from bucketed_scene_flow_eval.datasets.waymoopen import (
+    WaymoOpenCausalSceneFlow,
+    WaymoOpenNonCausalSceneFlow,
+)
+from bucketed_scene_flow_eval.interfaces import AbstractDataset
 
-importable_classes = [Argoverse2SceneFlow, WaymoOpenSceneFlow]
+importable_classes = [
+    Argoverse2CausalSceneFlow,
+    Argoverse2NonCausalSceneFlow,
+    WaymoOpenCausalSceneFlow,
+    WaymoOpenNonCausalSceneFlow,
+]
 name_to_class_lookup = {cls.__name__.lower(): cls for cls in importable_classes}
 
 
-def construct_dataset(name: str, args: dict):
+def construct_dataset(name: str, args: dict) -> AbstractDataset:
     name = name.lower()
     if name not in name_to_class_lookup:
         raise ValueError(f"Unknown dataset name: {name}")

bucketed_scene_flow_eval/datasets/argoverse2/__init__.py

@@ -1,16 +1,24 @@
-from .argoverse_raw_data import ArgoverseRawSequence, ArgoverseRawSequenceLoader
+from .argoverse_raw_data import (
+    DEFAULT_POINT_CLOUD_RANGE,
+    ArgoverseRawSequence,
+    ArgoverseRawSequenceLoader,
+    PointCloudRange,
+)
 from .argoverse_scene_flow import (
     ArgoverseNoFlowSequence,
     ArgoverseNoFlowSequenceLoader,
     ArgoverseSceneFlowSequence,
     ArgoverseSceneFlowSequenceLoader,
 )
-from .dataset import Argoverse2SceneFlow
+from .dataset import Argoverse2CausalSceneFlow, Argoverse2NonCausalSceneFlow
 
 __all__ = [
-    "Argoverse2SceneFlow",
-    "ArgoverseRawSequenceLoader",
+    "Argoverse2CausalSceneFlow",
+    "Argoverse2NonCausalSceneFlow",
+    "ArgoverseNoFlowSequence",
+    "ArgoverseNoFlowSequenceLoader",
     "ArgoverseRawSequence",
-    "ArgoverseSceneFlowSequenceLoader",
+    "ArgoverseRawSequenceLoader",
     "ArgoverseSceneFlowSequence",
+    "ArgoverseSceneFlowSequenceLoader",
 ]