[WIP] Instanseg inference

scripts/convert_geff_to_db.py

@@ -0,0 +1,95 @@
+"""
+Script to convert GEFF files to Ultrack SQLite databases for the cellmotv1 tracking challenge.
+
+For each experiment folder:
+  - Runs: trackedit convert geff-to-db <experiment>.geff -o data.db
+  - Writes: metadata.toml with shape = [ 100, 64, 256, 256,]
+
+Usage:
+  python convert_geff_to_db.py              # test mode: processes one experiment, output to Desktop
+  python convert_geff_to_db.py --all        # processes all experiments, output next to geff/zarr
+"""
+
+import subprocess
+import sys
+from pathlib import Path
+
+BASE_DIR = Path(
+    "/hpc/projects/group.royer/people/thibaut.goldsborough/tracking-challenge/cellmotv1"
+)
+
+# Used only in test mode (no write access to BASE_DIR)
+TEST_OUTPUT_DIR = Path("/home/teun.huijben/Desktop/tracking-challenge/cellmotv1")
+
+SHAPE = [100, 64, 256, 256]
+
+METADATA_CONTENT = f"shape = {SHAPE}\n"
+
+
+def process_experiment(experiment_dir: Path, out_dir: Path) -> bool:
+    geff_files = list(experiment_dir.glob("*.geff"))
+    if not geff_files:
+        print(f"  [SKIP] No .geff file found in {experiment_dir}")
+        return False
+
+    geff_path = geff_files[0]
+    out_dir.mkdir(parents=True, exist_ok=True)
+    db_output = out_dir / "data.db"
+    metadata_path = out_dir / "metadata.toml"
+
+    print(f"  Converting: {geff_path.name}")
+    result = subprocess.run(
+        ["trackedit", "convert", "geff-to-db", str(geff_path), "-o", str(db_output)],
+        capture_output=True,
+        text=True,
+    )
+
+    if result.returncode != 0:
+        print(f"  [ERROR] Conversion failed:\n{result.stderr}")
+        return False
+
+    print(f"  Written:    {db_output}")
+
+    metadata_path.write_text(METADATA_CONTENT)
+    print(f"  Written:    {metadata_path}")
+
+    return True
+
+
+def main():
+    run_all = "--all" in sys.argv
+
+    date_folders = sorted(BASE_DIR.iterdir())
+
+    if run_all:
+        experiments = [
+            exp
+            for date_folder in date_folders
+            for exp in sorted(date_folder.iterdir())
+            if exp.is_dir()
+        ]
+        print(f"Processing all {len(experiments)} experiments...\n")
+        pairs = [(exp, exp) for exp in experiments]
+    else:
+        # Test mode: first experiment of the first folder, write to Desktop
+        first_folder = date_folders[0]
+        exp = sorted(first_folder.iterdir())[0]
+        experiments = [exp]
+        relative = exp.relative_to(BASE_DIR)
+        out = TEST_OUTPUT_DIR / relative
+        pairs = [(exp, out)]
+        print("TEST MODE: processing 1 experiment")
+        print(f"  Input:  {exp}")
+        print(f"  Output: {out}\n")
+
+    success = 0
+    for exp, out in pairs:
+        print(f"[{exp.parent.name}/{exp.name}]")
+        if process_experiment(exp, out):
+            success += 1
+
+    print(f"\nDone: {success}/{len(experiments)} experiments converted successfully.")
+
+
+if __name__ == "__main__":
+    main()

scripts/script_instanseg.py

@@ -0,0 +1,70 @@
+import sys
+import warnings
+from pathlib import Path
+
+import numpy as np
+
+from trackedit.run import run_trackedit
+
+# Databases saved with numpy>2 need np._core.numeric, which is not available in numpy<2, hence the following hack
+sys.modules["numpy._core.numeric"] = np.core.numeric
+
+warnings.filterwarnings("ignore", category=FutureWarning, message=".*qt_viewer.*")
+
+# **********INPUTS*********
+# path to the working directory that contains the database file AND metadata.toml:
+working_directory = Path(
+    "/hpc/projects/group.royer/people/teun.huijben/data/Thibault/4th_exp//masks_on_geff/"
+)
+# name of the database file to start from, or "latest" to start from the latest version, defaults to "data.db"
+db_filename_start = "latest"
+# maximum number of frames display, defaults to None (use all frames)
+tmax = 100
+# (Z),Y,X, defaults to (1, 1, 1)
+scale = (1.625, 0.40625, 0.40625)
+# overwrite existing database/changelog, defaults to False (not used when db_filename_start is "latest")
+allow_overwrite = False
+
+# OPTIONAL: imaging data
+imaging_zarr_file = (
+    "/hpc/projects/group.royer/people/teun.huijben/data/Thibault/4th_exp/first_fov.zarr"
+)
+imaging_channel = "0"
+imaging_layer_names = ["dense", "sparse"]
+
+# OPTIONAL: annotation mapping (default is neuromast cell types)
+# annotation_mapping = {
+#     1: {"name": "hair", "color": [0.0, 1.0, 0.0, 1.0]},  # green
+#     2: {"name": "support", "color": [1.0, 0.1, 0.6, 1.0]},  # pink
+#     3: {"name": "mantle", "color": [0.0, 0.0, 0.9, 1.0]},  # blue
+# }
+annotation_mapping = None
+
+# OPTIONAL: InstanSeg model for interactive cell segmentation
+# Enable this to add cells via InstanSeg inference instead of spherical masks
+flag_allow_adding_instanseg_cell = True
+instanseg_model_path = (
+    "/hpc/projects/group.royer/people/teun.huijben/data/Thibault/model_96.pt"
+)
+instanseg_device = None  # 'cuda', 'cpu', or None for auto-detect
+# *************************
+
+if __name__ == "__main__":
+    run_trackedit(
+        working_directory=working_directory,
+        db_filename=db_filename_start,
+        tmax=tmax,
+        scale=scale,
+        allow_overwrite=allow_overwrite,
+        imaging_zarr_file=imaging_zarr_file,
+        imaging_channel=imaging_channel,
+        imaging_layer_names=imaging_layer_names,
+        annotation_mapping=annotation_mapping,
+        flag_allow_adding_spherical_cell=True,
+        adding_spherical_cell_radius=10,
+        flag_remove_red_flags_at_edge=True,
+        remove_red_flags_at_edge_threshold=10,
+        flag_allow_adding_instanseg_cell=flag_allow_adding_instanseg_cell,
+        instanseg_model_path=instanseg_model_path,
+        instanseg_device=instanseg_device,
+    )

trackedit/DatabaseHandler.py

@@ -28,7 +28,9 @@
     combine_red_flags,
     filter_red_flags_at_edge,
     find_all_starts_and_ends,
+    find_area_changes,
     find_overlapping_cells,
+    find_trajectory_changes,
 )
 from trackedit.utils.utils import (
     annotations_to_zarr,
@@ -605,6 +607,7 @@ def db_to_df(
                 NodeDB.z,
                 NodeDB.y,
                 NodeDB.x,
+                NodeDB.area,
                 NodeDB.generic,
             ),
         )
@@ -714,8 +717,16 @@ def find_all_red_flags(self) -> pd.DataFrame:
             df, self.db_path_new
         )
 
+        # Trajectory changes (jumps and direction changes)
+        rfs_trajectory = find_trajectory_changes(df, self.scale)
+
+        # Area/volume changes
+        rfs_area = find_area_changes(df)
+
         # Combine all red flag detection results
-        result_df = combine_red_flags(rfs_starts_and_ends, rfs_overlap)
+        result_df = combine_red_flags(
+            rfs_starts_and_ends, rfs_overlap, rfs_trajectory, rfs_area
+        )
 
         # ToDo: make option to filter redflags in the first two timepoints
         # (useful for neuromast with suboptimal beginning)
@@ -729,6 +740,7 @@ def find_all_red_flags(self) -> pd.DataFrame:
                 data_shape=self.data_shape_full[1:],
                 edge_threshold=self.remove_red_flags_at_edge_threshold,
                 ndim=self.ndim,
+                scale=self.scale,
             )
 
         return result_df