Add scripts for evaluating automatic tracking results

scripts/measure_for_automatic.py

@@ -0,0 +1,88 @@
+import os
+
+import json
+import h5py
+import numpy as np
+import pandas as pd
+import imageio.v3 as imageio
+
+from deepcell_tracking import isbi_utils
+
+from traccuracy import run_metrics
+from traccuracy.matchers import CTCMatcher
+from traccuracy._tracking_graph import TrackingGraph
+from traccuracy.metrics import CTCMetrics, DivisionMetrics
+from traccuracy.loaders._ctc import _get_node_attributes, ctc_to_graph, _check_ctc
+
+
+ROOT = "/scratch/share/cidas/cca/experiments/results_for_automatic_tracking"
+
+
+def _get_tracks_to_isbi():
+    with open(os.path.join(ROOT, "tracking_result.json"), "r") as f:
+        data = json.load(f)
+
+    segmentation = imageio.imread(os.path.join(ROOT, "tracking_result.tif"))
+
+    # we convert the parent ids to frames
+    track_info = {}
+    for parent_id, daughter_ids in data.items():
+        parent_id = int(parent_id)
+        frames = np.where(segmentation == parent_id)[0]
+
+        # store the parent's track information
+        track_info[parent_id] = {
+            'frames': list(np.unique(frames)),
+            'daughters': list(daughter_ids),
+            'frame_div': frames.max() + 1,
+            'parent': None,
+            'label': parent_id,
+        }
+
+        # next, store the daughter's track information
+        for daughter_id in daughter_ids:
+            frames = np.where(segmentation == daughter_id)[0]
+            track_info[daughter_id] = {
+                'frames': list(np.unique(frames)),
+                'daughters': [],
+                'frame_div': None,
+                'parent': parent_id,
+                'label': daughter_id
+            }
+
+    track_df = isbi_utils.trk_to_isbi(track_info)
+    return segmentation, track_df
+
+
+def _get_metrics_for_autotrack(segmentation, seg_df):
+    # NOTE: for ground-truth
+    with h5py.File(os.path.join(ROOT, "tracking_micro_sam.h5")) as f:
+        gt = f['labels'][:]
+
+    gt_nodes = _get_node_attributes(gt)
+    gt_df = pd.read_csv(os.path.join(ROOT, "gt_tracks.csv"))
+    gt_G = ctc_to_graph(gt_df, gt_nodes)
+    _check_ctc(gt_df, gt_nodes, gt)
+    gt_T = TrackingGraph(gt_G, segmentation=gt, name="DynamicNuclearNet-GT")
+
+    # NOTE: for segmentation results
+    # calculate node attributes for each detection
+    seg_nodes = _get_node_attributes(segmentation)
+    seg_G = ctc_to_graph(seg_df, seg_nodes)
+    seg_T = TrackingGraph(seg_G, segmentation=segmentation, name="DynamicNuclearNet-autotracking")
+
+    ctc_results = run_metrics(
+        gt_data=gt_T,
+        pred_data=seg_T,
+        matcher=CTCMatcher(),
+        metrics=[CTCMetrics(), DivisionMetrics(max_frame_buffer=0)]
+    )
+    print(ctc_results)
+
+
+def main():
+    segmentation, seg_df = _get_tracks_to_isbi()
+    _get_metrics_for_autotrack(segmentation, seg_df)
+
+
+main()