Release 1.3.1

CHANGELOG.MD

@@ -5,6 +5,20 @@
 [//]: # (this file, may be copied, modified, propagated, or distributed except according to)
 [//]: # (the terms contained in the file 'LICENCE.txt'.)
 
+## [v1.3.1] – 2026-05-11
+
+### Fixed
+- **Cutana 0.3.0 compatibility** by removing the obsolete `synchronised_loading` kwarg from `StreamingOrchestrator.init_streaming` calls
+- **Pretrained weight loading** when the repository is cloned without `git-lfs`: fall back to a HuggingFace download instead of failing on LFS pointer files (#7)
+
+### Changed
+- **Security:** replaced pickle-based `torch.save` / `torch.load` checkpointing with `safetensors`, removing the pickle deserialisation risk on untrusted model files (#9)
+- **PyTorch minimum** raised to `>=2.6` in the conda environments to pick up upstream CVE fixes (#11)
+- **Cutana minimum** raised to `>=0.3.0` in `environment.yml`, `environment_CI.yml`, and `pyproject.toml`
+
+### Documentation
+- **Programmatic Evaluation (Headless Mode)** section added to the README, showing how to run batch predictions through `Session.load_model()` + `Session.evaluate_all_images()` without launching the UI (#10)
+
 ## [v1.3.0] – 2026-02-11
 
 ### Added

anomaly_match/__init__.py

@@ -12,7 +12,7 @@
 from .utils.print_cfg import print_cfg
 from .utils.set_log_level import set_log_level
 
-__version__ = "1.3.0"
+__version__ = "1.3.1"
 
 __all__ = [
     "get_default_cfg",

environment.yml

@@ -33,7 +33,7 @@ dependencies:
   - pip
   - pip:
       - albumentations
-      - cutana>=0.2.1
+      - cutana>=0.3.0
       - fitsbolt>=0.2
       - opencv-python-headless
       - safetensors

environment_CI.yml

@@ -38,4 +38,4 @@ dependencies:
       - safetensors
       - timm
       - fitsbolt>=0.2
-      - cutana>=0.2.1
+      - cutana>=0.3.0

prediction_process_cutana.py

@@ -210,9 +210,7 @@ def evaluate_images_from_cutana(
 
         cutana_orchestrator = cutana.StreamingOrchestrator(cutana_config)
 
-        cutana_orchestrator.init_streaming(
-            batch_size=batch_size, write_to_disk=False, synchronised_loading=False
-        )
+        cutana_orchestrator.init_streaming(batch_size=batch_size, write_to_disk=False)
     except Exception as e:
         logger.error(f"Failed to initialize Cutana orchestrator: {e}")
         raise
@@ -266,7 +264,11 @@ def evaluate_images_from_cutana(
         # Handle empty batches (cutana returns [] if all cutouts failed)
         if isinstance(batch_data, list):
             if len(batch_data) == 0:
-                logger.warning(f"Batch {batch_idx} returned empty cutouts (list), skipping")
+                logger.warning(
+                    f"Batch {batch_idx} returned empty cutouts (list), skipping. "
+                    f"Please ensure that number of extensions in Cutana source "
+                    f"catalogues matches those of the training set."
+                )
                 continue
             # Convert list to numpy array if needed
             batch_data = np.array(batch_data)

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "anomaly_match"
-version = "1.3.0"
+version = "1.3.1"
 description = "A tool for anomaly detection in images using semi-supervised and active learning with a GUI"
 readme = "README.md"
 license = { file = "LICENSE.txt" }
@@ -32,7 +32,7 @@ classifiers = [
 dependencies = [
     "albumentations",
     "astropy",
-    "cutana>=0.2.1",
+    "cutana>=0.3.0",
     "dotmap",
     "fitsbolt>=0.2",
     "h5py",

tests/e2e/test_normalisation_consistency.py

@@ -86,7 +86,7 @@ def _run_cutana_normalised(csv_path, fitsbolt_cfg, n_output_channels=3):
     config.external_fitsbolt_cfg = fitsbolt_cfg
 
     orchestrator = cutana.StreamingOrchestrator(config)
-    orchestrator.init_streaming(batch_size=10, write_to_disk=False, synchronised_loading=False)
+    orchestrator.init_streaming(batch_size=10, write_to_disk=False)
 
     all_cutouts = []
     for _ in range(orchestrator.get_batch_count()):
@@ -112,7 +112,7 @@ def _extract_raw_cutouts(csv_path, output_dir):
     config.output_dir = output_dir
 
     orchestrator = cutana.StreamingOrchestrator(config)
-    orchestrator.init_streaming(batch_size=10, write_to_disk=True, synchronised_loading=False)
+    orchestrator.init_streaming(batch_size=10, write_to_disk=True)
     for _ in range(orchestrator.get_batch_count()):
         orchestrator.next_batch()
     orchestrator.cleanup()

tests/unit/test_checkpoint_io.py

@@ -7,11 +7,14 @@
 
 """Unit tests for checkpoint_io: safetensors-based model checkpoint serialization."""
 
+import json
+
 import numpy as np
 import pytest
 import torch
 from dotmap import DotMap
 from fitsbolt.normalisation.NormalisationMethod import NormalisationMethod
+from safetensors import safe_open
 
 from anomaly_match.data_io.checkpoint_io import load_checkpoint, save_checkpoint
 
@@ -228,10 +231,6 @@ def test_no_pickle_in_file(self, tmp_path):
 
     def test_metadata_is_plain_json(self, tmp_path):
         """All metadata in the safetensors header is valid JSON strings."""
-        import json
-
-        from safetensors import safe_open
-
         path = save_checkpoint(_make_full_checkpoint(), tmp_path / "model")
         with safe_open(str(path), framework="pt") as f:
             metadata = f.metadata()