Migrate to pathlib.Path instead of str (first phase) (#3350)

* fix path resolution on network drives. Replace Path(..).resolve() with Path(..).absolute() * Add safe_resolve helper for path resolution * replace str with pathlib.Path * fix redundant resolution of resolved path in memory_replay.py * fix open -> os.stat * replace os.stat -> open for files, listdir for dirs. * fix TestReadConfigProjectPath * Update deeplabcut/utils/auxiliaryfunctions.py * additional replacements: str ->Path * Update function signatures str -> str | Path * Add ruff linting rule PTH * Open dataset file directly using Path.open Fix path handling when loading the dataset pickle. The code previously attempted Path(project_path) / file_name even though file_name already included the project path, resulting in an incorrect path. This changes the context manager to call file_name.open('rb') directly, removing the redundant join and preventing path errors when loading the pickle. * Use context manager when writing pose_cfg.yaml Replace direct ruamel_file.dump(...open('w')) call with a with-statement to open pose_cfg.yaml. This ensures the file handle is properly closed after writing and avoids potential resource leaks. * Pass string path to cv2.VideoCapture Convert video_path to str when constructing cv2.VideoCapture to ensure pathlib.Path objects are handled correctly and avoid type errors when opening videos. This makes VideoReader more robust when given Path instances. * Use explicit loop to unlink frame files Replace the list-comprehension side-effect used to delete temporary frame files with an explicit for-loop that checks p.is_file() before calling unlink. This avoids building an unused list, prevents errors from trying to unlink non-file entries (e.g., directories), and improves readability and safety. * Fix CircleCI failure * fix Matplotlib Path import shadowing ->MPLPath * use pathlib consistently in `metrics` * use safe_resolve in export * remove redundant defensive block for windows paths on python 3.6 * remove unused `get_immediate_subdirectories` * use Path in plot_edge_affinity_distributions * remove PTH enforcement in linting * use safe_resolve instead of resolve in `training` and `auxiliaryfunctions` * Update GUI assets import (#3370) * Add GUI asset helper utilities Introduce deeplabcut.gui.gui_assets module providing utilities to load bundled GUI assets. It locates the package assets directory via importlib.resources, and exposes resource_bytes, resource_text, get_assets_dir, get_style_qss, pixmap_from_resource and icon_from_resource helpers (using PySide6 QPixmap/QIcon). Includes error handling when image data cannot be loaded and uses type annotations for clarity. * Use gui_assets for icons and pixmaps Replace direct filesystem asset loading (BASE_DIR/assets and manual Path lookups) with centralized gui_assets helpers (icon_from_resource, pixmap_from_resource, get_style_qss/get_assets_dir) across multiple GUI modules. Updated imports and calls in components, launch_script, create_project, modelzoo, open_project, train_network, and window to use the new resource functions and remove redundant logo path handling, improving resource access and packaging robustness. * fix path concatenation in `make_labeled_images_from_dataframe` --------- Co-authored-by: Jaap de Ruyter van Steveninck <32810691+deruyter92@users.noreply.github.com> * Adjust config path tests for Path objects Update `test_auxiliaryfunctions.py` to assert `project_path` as a `pathlib.Path` instead of a string. The tests now cast to `str` only for the `"Volume{"` check, use `.exists()` for filesystem validation, and compare against `project_dir.absolute()` directly. * Fix os import in safe_resolve Import `os` inside `safe_resolve` before calling `os.listdir` so directory resolution no longer risks a `NameError`. This also removes an unused `os` import in `test_auxiliaryfunctions.py` and a stray blank line in `video_processor.py`. * address remaining `.resolve()` cases: prefer `.absolute()`, or otherwise `safe_resolve()` * update `safe_resolve`: add logging, use os.stat (more efficient) * update test safe_resolve: allow fallback and test open/closing instead Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> * fix remaining Path.resolve() * fix reintroduced pathlib `resolve()` occurences (see #3349) * use pathlib Path in GUI codebase * use pathlib Path in memory replay * GUI rename field `config` to `config_path` * GUI use os.fspath for str conversions where required * use pathlib Path in create_project * use pathlib Path across pose_estimation_pytorch internally * use centralized yaml dumper (which normalizes Path -> str) * fix materialize.py downstream old str.split -> Path * fix testsscripts str usage * fix old str usage in trainingsetmanipulation * allow str type at TF boundary * allow str formatting in tensorflow code * fix config update in train_from_coco.py PoseConfig.model.backbone is currently a dict-type but freeze_bn_stats were set as attribute. This commit fixes that and treats backbone correctly as dict. --------- Co-authored-by: Cyril Achard <cyril.achard@epfl.ch> Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
C-Achard · Jul 16, 2026 · Jul 15, 2026 · Jul 15, 2026 · Jul 15, 2026 · Jul 15, 2026
commit 30a8155d4d27984b272cedcf4a7b0a45b5951a42
diff --git a/deeplabcut/benchmark/__init__.py b/deeplabcut/benchmark/__init__.py
@@ -13,12 +13,13 @@
 import json
 import os
 from collections.abc import Container
+from pathlib import Path
 from typing import Literal
 
 from deeplabcut.benchmark.base import Benchmark, Result, ResultCollection
 
-DATA_ROOT = os.path.join(os.getcwd(), "data")
-CACHE = os.path.join(os.getcwd(), ".results")
+DATA_ROOT = Path.cwd() / "data"
+CACHE = Path.cwd() / ".results"
 
 __registry = []
 
@@ -99,20 +100,20 @@ def evaluate(
 
 
 def get_filepath(basename: str):
-    return os.path.join(DATA_ROOT, basename)
+    return DATA_ROOT / basename
 
 
 def savecache(results: ResultCollection):
-    with open(CACHE, "w") as fh:
+    with Path(CACHE).open("w") as fh:
         json.dump(results.todicts(), fh, indent=2)
 
 
 def loadcache(cache=CACHE, on_missing: Literal["raise", "ignore"] = "ignore") -> ResultCollection:
-    if not os.path.exists(cache):
+    if not Path(cache).exists():
         if on_missing == "raise":
             raise FileNotFoundError(cache)
         return ResultCollection()
-    with open(cache) as fh:
+    with Path(cache).open() as fh:
         try:
             data = json.load(fh)
         except json.decoder.JSONDecodeError as e:

diff --git a/deeplabcut/benchmark/metrics.py b/deeplabcut/benchmark/metrics.py
@@ -11,9 +11,9 @@
 
 """Evaluation metrics for the DeepLabCut benchmark."""
 
-import os
 import pickle
 from collections import defaultdict
+from pathlib import Path
 
 import numpy as np
 import pandas as pd
@@ -33,7 +33,7 @@ def _format_gt_data(h5file: str, test_indices: list[int] | None = None):
     except KeyError:
         n_unique = 0
     guarantee_multiindex_rows(df)
-    file_paths = [os.path.join(*row) for row in df.index.to_list()]
+    file_paths = [Path(*row) for row in df.index.to_list()]
     temp = (
         df.stack("individuals", dropna=False)
         .reindex(animals, level="individuals")
@@ -98,19 +98,18 @@ def calc_prediction_errors(preds, gt):
     return errors
 
 
-def _map(strings, substrings):
+def _map(paths: list[Path], subpaths: list[Path]) -> dict[Path, Path]:
     """Map image paths from predicted data to GT as the first are typically absolute
     whereas the latter are relative to the project path."""
-
-    lookup = dict()
-    strings_ = strings.copy()
-    substrings_ = substrings.copy()
-    while strings_:
-        string = strings_.pop()
-        for s in substrings_:
-            if string.endswith(s):
-                lookup[string] = s
-                substrings_.remove(s)
+    lookup = {}
+    paths_ = paths.copy()
+    subpaths_ = subpaths.copy()
+    while paths_:
+        path = paths_.pop()
+        for s in subpaths_:
+            if path.parts[-len(s.parts) :] == s.parts:
+                lookup[path] = s
+                subpaths_.remove(s)
                 break
     return lookup
 
@@ -146,6 +145,7 @@ def calc_map_from_obj(
     drop_kpts=None,
 ):
     """Calculate mean average precision (mAP) based on predictions."""
+    eval_results = {Path(k): v for k, v in eval_results_obj.items()}
     df = pd.read_hdf(h5_file)
     try:
         df.drop("single", level="individuals", axis=1, inplace=True)
@@ -157,7 +157,7 @@ def calc_map_from_obj(
     test_indices = _load_test_indices(metadata_file)
     df_test = df.iloc[test_indices]
     test_images = load_test_images(h5_file, metadata_file)
-    missing_images = set(test_images) - set(eval_results_obj.keys())
+    missing_images = set(test_images) - set(eval_results)
     if len(missing_images) > 0:
         raise ValueError(
             f"Failed to compute the test mAP: there are test images missing from theprediction object: {missing_images}"
@@ -184,7 +184,7 @@ def calc_map_from_obj(
         for ind in sorted(drop_kpts, reverse=True):
             kpts.pop(ind)
 
-    assemblies_pred = conv_obj_to_assemblies(eval_results_obj, kpts)
+    assemblies_pred = conv_obj_to_assemblies(eval_results, kpts)
     with deeplabcut.benchmark.utils.DisableOutput():
         oks = inferenceutils.evaluate_assembly(
             assemblies_pred,
@@ -213,7 +213,7 @@ def calc_rmse_from_obj(
         for ind in sorted(drop_kpts, reverse=True):
             kpts.pop(ind)
 
-    test_objects = {k: v for k, v in eval_results_obj.items() if k in gt["annotations"].keys()}
+    test_objects = {Path(k): v for k, v in eval_results_obj.items() if Path(k) in gt["annotations"]}
     if len(gt["annotations"]) != len(test_objects):
         gt_images = list(gt["annotations"].keys())
         missing_images = [img for img in gt_images if img not in test_objects]
@@ -239,22 +239,17 @@ def calc_rmse_from_obj(
     return np.nanmean(errors[..., 0])
 
 
-def load_test_images(h5file: str, metadata: str) -> list[str]:
+def load_test_images(h5file: str, metadata: str) -> list[Path]:
     """Returns the names of the test images for the benchmark, in the order
     corresponding to the test indices."""
     df = pd.read_hdf(h5file)
     test_indices = _load_test_indices(metadata)
     df_test = df.iloc[test_indices]
-    test_images = []
-    for img_path in df_test.index:
-        if not isinstance(img_path, str):
-            img_path = os.path.join(*img_path)
-        test_images.append(img_path)
-    return test_images
+    return [Path(img_path) if isinstance(img_path, str) else Path(*img_path) for img_path in df_test.index]
 
 
 def _load_test_indices(shuffle_metadata_path: str) -> list[int]:
     """Returns the indices of test images in the training dataset dataframe."""
-    with open(shuffle_metadata_path, "rb") as f:
+    with Path(shuffle_metadata_path).open("rb") as f:
         test_indices = set([int(i) for i in pickle.load(f)[2]])
     return list(sorted(test_indices))
diff --git a/deeplabcut/benchmark/utils.py b/deeplabcut/benchmark/utils.py
@@ -18,6 +18,7 @@
 import os
 import pkgutil
 import sys
+from pathlib import Path
 
 
 class RedirectStdStreams:
@@ -46,7 +47,7 @@ def __exit__(self, exc_type, exc_value, traceback):
 
 class DisableOutput(RedirectStdStreams):
     def __init__(self):
-        devnull = open(os.devnull, "w")
+        devnull = Path(os.devnull).open("w")
         super().__init__(stdout=devnull, stderr=devnull)