Add typing fixes. (#239)

* Add typing fixes.

* Revert ignore comments.

* Fix rust lint.

* Fix rust lint.

* Update environment.

* Add libtiff.

* Update channels.

* Try torch assert fix.

* Update python versions.

* Fix 3.11 noop.

* Fix 3.11 ci.

* Remove excess dep.

* Remove excess dep.

* Fix constraint.

* Add pytorch channel.

* Remove redundant python req + uncomment code.

.github/workflows/ci.yaml

@@ -14,7 +14,7 @@ jobs:
       matrix:
         os: [macos-latest, ubuntu-latest]
         python_version:
-          ["3.8", "3.9", "3.10"]
+          ["3.9", "3.10", "3.11"]
     defaults:
       run:
         shell: bash -l {0}

conda/environment.yml

@@ -27,4 +27,4 @@ dependencies:
   - tqdm
   - universal_pathlib
   - pip:
-    - git+https://github.com/JonathonLuiten/TrackEval.git
+      - git+https://github.com/JonathonLuiten/TrackEval.git

rust/src/data_loader.rs

@@ -157,7 +157,7 @@ impl DataLoader {
     pub fn get(&self, index: usize) -> Sweep {
         let row = self.file_index.0.get_row(index).unwrap().0;
         let (log_id, timestamp_ns) = (
-            row.get(0).unwrap().get_str().unwrap(),
+            row.first().unwrap().get_str().unwrap(),
             row.get(1).unwrap().try_extract::<u64>().unwrap(),
         );
 
@@ -313,7 +313,7 @@ impl DataLoader {
     pub fn get_synchronized_images(&self, index: usize) -> Vec<TimeStampedImage> {
         let row = self.file_index.0.get_row(index).unwrap().0;
         let (log_id, _) = (
-            row.get(0).unwrap().get_str().unwrap(),
+            row.first().unwrap().get_str().unwrap(),
             row.get(1).unwrap().try_extract::<u64>().unwrap(),
         );
 

src/av2/torch/structures/flow.py

@@ -97,7 +97,7 @@ def from_sweep_pair(cls, sweeps: Tuple[Sweep, Sweep]) -> Flow:
                 c1_SE3_c0 = c1.dst_SE3_object.compose(c0.dst_SE3_object.inverse())
                 flow[obj_mask] = (
                     torch.as_tensor(
-                        c1_SE3_c0.transform_point_cloud(obj_pts.numpy()),
+                        c1_SE3_c0.transform_point_cloud(obj_pts_npy),
                         dtype=torch.float32,
                     )
                     - obj_pts

src/av2/utils/dense_grid_interpolation.py

@@ -45,13 +45,13 @@ def interp_dense_grid_from_sparse(
     """
     if interp_method not in ["linear", "nearest"]:
         raise ValueError("Unknown interpolation method.")
+    if grid_img.dtype != values.dtype:
+        raise ValueError("Grid and values should be the same datatype.")
 
     if points.shape[0] < MIN_REQUIRED_POINTS_SIMPLEX:
         # return the empty grid, since we can't interpolate.
         return grid_img
 
-    output_dtype = values.dtype
-
     # get (x,y) tuples back
     grid_coords = mesh_grid.get_mesh_grid_as_point_cloud(
         min_x=0, max_x=grid_w - 1, min_y=0, max_y=grid_h - 1
@@ -65,4 +65,4 @@ def interp_dense_grid_from_sparse(
     v = grid_coords[:, 1].astype(np.int32)
     # Now index in at (y,x) locations
     grid_img[v, u] = interp_vals
-    return grid_img.astype(output_dtype)
+    return grid_img

src/av2/utils/synchronization_database.py

@@ -142,12 +142,14 @@ def __init__(
             self.per_log_cam_timestamps_index[log_id] = {}
             for cam_name in list(RingCameras) + list(StereoCameras):
                 sensor_folder_wildcard = (
-                    f"{dataset_dir}/{log_id}/sensors/cameras/{cam_name}/*.jpg"
+                    f"{dataset_dir}/{log_id}/sensors/cameras/{cam_name.value}/*.jpg"
                 )
                 cam_timestamps = get_timestamps_from_sensor_folder(
                     sensor_folder_wildcard
                 )
-                self.per_log_cam_timestamps_index[log_id][cam_name] = cam_timestamps
+                self.per_log_cam_timestamps_index[log_id][
+                    cam_name.value
+                ] = cam_timestamps
 
             sensor_folder_wildcard = f"{dataset_dir}/{log_id}/sensors/lidar/*.feather"
             lidar_timestamps = get_timestamps_from_sensor_folder(sensor_folder_wildcard)

tests/unit/datasets/sensor/test_sensor_dataloader.py

@@ -24,33 +24,33 @@
 
 def _create_dummy_sensor_dataloader(log_id: str) -> SensorDataloader:
     """Create a dummy sensor dataloader."""
-    with Path(tempfile.TemporaryDirectory().name) as sensor_dataset_dir:
-        for sensor_name, timestamps_ms in SENSOR_TIMESTAMPS_MS_DICT.items():
-            for t in timestamps_ms:
-                if "ring" in sensor_name:
-                    fpath = Path(
-                        sensor_dataset_dir,
-                        "dummy",
-                        log_id,
-                        "sensors",
-                        "cameras",
-                        sensor_name,
-                        f"{int(t*1e6)}.jpg",
-                    )
-                    Path(fpath).parent.mkdir(exist_ok=True, parents=True)
-                    fpath.open("w").close()
-                elif "lidar" in sensor_name:
-                    fpath = Path(
-                        sensor_dataset_dir,
-                        "dummy",
-                        log_id,
-                        "sensors",
-                        sensor_name,
-                        f"{int(t*1e6)}.feather",
-                    )
-                    Path(fpath).parent.mkdir(exist_ok=True, parents=True)
-                    fpath.open("w").close()
-        return SensorDataloader(dataset_dir=sensor_dataset_dir, with_cache=False)
+    sensor_dataset_dir = Path(tempfile.TemporaryDirectory().name)
+    for sensor_name, timestamps_ms in SENSOR_TIMESTAMPS_MS_DICT.items():
+        for t in timestamps_ms:
+            if "ring" in sensor_name:
+                fpath = Path(
+                    sensor_dataset_dir,
+                    "dummy",
+                    log_id,
+                    "sensors",
+                    "cameras",
+                    sensor_name,
+                    f"{int(t*1e6)}.jpg",
+                )
+                Path(fpath).parent.mkdir(exist_ok=True, parents=True)
+                fpath.open("w").close()
+            elif "lidar" in sensor_name:
+                fpath = Path(
+                    sensor_dataset_dir,
+                    "dummy",
+                    log_id,
+                    "sensors",
+                    sensor_name,
+                    f"{int(t*1e6)}.feather",
+                )
+                Path(fpath).parent.mkdir(exist_ok=True, parents=True)
+                fpath.open("w").close()
+    return SensorDataloader(dataset_dir=sensor_dataset_dir, with_cache=False)
 
 
 def test_sensor_data_loader_milliseconds() -> None:

tests/unit/evaluation/scene_flow/test_sf_eval.py

@@ -271,67 +271,65 @@ def test_average_metrics() -> None:
 
     Verify that the weighted average metric breakdown has the correct subset counts and values.
     """
-    with Path(tempfile.TemporaryDirectory().name) as test_dir:
-        test_dir.mkdir()
-        anno_dir = test_dir / "annotations"
-        anno_dir.mkdir()
-
-        pred_dir = test_dir / "predictions"
-        pred_dir.mkdir()
-
-        timestamp_ns_1 = 111111111111111111
-        timestamp_ns_2 = 111111111111111112
-
-        write_annotation(
-            gts_classes,
-            gts_close,
-            gts_dynamic,
-            gts_valid,
-            gts,
-            ("log", timestamp_ns_1),
-            anno_dir,
-        )
-        write_annotation(
-            gts_classes,
-            gts_close,
-            gts_dynamic,
-            gts_valid,
-            gts,
-            ("log", timestamp_ns_2),
-            anno_dir,
-        )
+    test_dir = Path(tempfile.TemporaryDirectory().name)
+    test_dir.mkdir()
+    anno_dir = test_dir / "annotations"
+    anno_dir.mkdir()
 
-        write_annotation(
-            gts_classes,
-            gts_close,
-            gts_dynamic,
-            gts_valid,
-            gts,
-            ("log_missing", timestamp_ns_1),
-            anno_dir,
-        )
+    pred_dir = test_dir / "predictions"
+    pred_dir.mkdir()
 
-        write_output_file(dts_perfect, dts_dynamic, ("log", timestamp_ns_1), pred_dir)
-        write_output_file(dts_perfect, dts_dynamic, ("log", timestamp_ns_2), pred_dir)
+    timestamp_ns_1 = 111111111111111111
+    timestamp_ns_2 = 111111111111111112
 
-        results_df = eval.evaluate_directories(anno_dir, pred_dir)
+    write_annotation(
+        gts_classes,
+        gts_close,
+        gts_dynamic,
+        gts_valid,
+        gts,
+        ("log", timestamp_ns_1),
+        anno_dir,
+    )
+    write_annotation(
+        gts_classes,
+        gts_close,
+        gts_dynamic,
+        gts_valid,
+        gts,
+        ("log", timestamp_ns_2),
+        anno_dir,
+    )
+
+    write_annotation(
+        gts_classes,
+        gts_close,
+        gts_dynamic,
+        gts_valid,
+        gts,
+        ("log_missing", timestamp_ns_1),
+        anno_dir,
+    )
 
-        assert len(results_df) == 16
-        assert results_df.Count.sum() == 18
+    write_output_file(dts_perfect, dts_dynamic, ("log", timestamp_ns_1), pred_dir)
+    write_output_file(dts_perfect, dts_dynamic, ("log", timestamp_ns_2), pred_dir)
 
-        assert np.allclose(results_df.EPE.mean(), 0.0)
-        assert np.allclose(results_df["ACCURACY_STRICT"].mean(), 1.0)
-        assert np.allclose(results_df["ACCURACY_RELAX"].mean(), 1.0)
-        assert np.allclose(results_df["ANGLE_ERROR"].mean(), 0.0)
+    results_df = eval.evaluate_directories(anno_dir, pred_dir)
 
-        assert results_df.TP.sum() == 2 * 2
-        assert results_df.TN.sum() == 2 * 2  # First true negative marked invalid
-        assert results_df.FP.sum() == 3 * 2
-        assert results_df.FN.sum() == 2 * 2
+    assert len(results_df) == 16
+    assert results_df.Count.sum() == 18
 
-        assert (
-            results_df.groupby(["Class", "Motion"]).Count.sum().Background.Dynamic == 0
-        )
+    assert np.allclose(results_df.EPE.mean(), 0.0)
+    assert np.allclose(results_df["ACCURACY_STRICT"].mean(), 1.0)
+    assert np.allclose(results_df["ACCURACY_RELAX"].mean(), 1.0)
+    assert np.allclose(results_df["ANGLE_ERROR"].mean(), 0.0)
+
+    assert results_df.TP.sum() == 2 * 2
+    assert results_df.TN.sum() == 2 * 2  # First true negative marked invalid
+    assert results_df.FP.sum() == 3 * 2
+    assert results_df.FN.sum() == 2 * 2
+
+    assert results_df.groupby(["Class", "Motion"]).Count.sum().Background.Dynamic == 0
     results_dict = eval.results_to_dict(results_df)
 
     assert len(results_dict) == 38