Add image segmentation support to GUI

perceptionmetrics/models/torch_detection.py

@@ -9,7 +9,10 @@
 import torch
 from torch.utils.data import DataLoader, Dataset
 from torchvision.transforms import v2 as transforms
-from torchvision import tv_tensors
+try:
+    from torchvision import tv_tensors
+except ImportError:
+    from torchvision import datapoints as tv_tensors
 from tqdm.auto import tqdm
 
 from perceptionmetrics.datasets import detection as detection_dataset
@@ -188,9 +191,14 @@ def __getitem__(
         # Convert boxes/labels to tensors
         if len(boxes) == 0:
             boxes = torch.zeros((0, 4), dtype=torch.float32)
-        boxes = tv_tensors.BoundingBoxes(
-            boxes, format="XYXY", canvas_size=(image.height, image.width)
-        )
+        if hasattr(tv_tensors, "BoundingBoxes"):
+            boxes = tv_tensors.BoundingBoxes(
+                boxes, format="XYXY", canvas_size=(image.height, image.width)
+            )
+        else:
+            boxes = tv_tensors.BoundingBox(
+                boxes, format="XYXY", spatial_size=(image.height, image.width)
+            )
         category_indices = torch.as_tensor(category_indices, dtype=torch.int64)
 
         target = {

perceptionmetrics/models/torch_segmentation.py

@@ -2,7 +2,7 @@
 import os
 import time
 import tempfile
-from typing import Any, List, Optional, Tuple, Union
+from typing import Any, Callable, List, Optional, Tuple, Union
 
 import numpy as np
 import pandas as pd
@@ -374,6 +374,8 @@ def eval(
         translation_direction: str = "dataset_to_model",
         predictions_outdir: Optional[str] = None,
         results_per_sample: bool = False,
+        progress_callback: Optional[Callable] = None,
+        metrics_callback: Optional[Callable] = None,
     ) -> pd.DataFrame:
         """Perform evaluation for an image segmentation dataset
 
@@ -389,6 +391,10 @@ def eval(
         :type predictions_outdir: Optional[str], optional
         :param results_per_sample: Whether to store results per sample or not, defaults to False. If True, predictions_outdir must be provided.
         :type results_per_sample: bool, optional
+        :param progress_callback: Optional callback called as progress_callback(processed, total), defaults to None
+        :type progress_callback: Optional[Callable], optional
+        :param metrics_callback: Optional callback called as metrics_callback(metrics_df, processed, total), defaults to None
+        :type metrics_callback: Optional[Callable], optional
         :return: DataFrame containing evaluation results
         :rtype: pd.DataFrame
         """
@@ -444,6 +450,9 @@ def eval(
 
         # Init metrics
         metrics_factory = um.SegmentationMetricsFactory(n_classes)
+        total_samples = len(dataset)
+        processed_samples = 0
+        evaluation_step = self.model_cfg.get("evaluation_step", 1)
 
         # Evaluation loop
         with torch.no_grad():
@@ -504,6 +513,26 @@ def eval(
                             os.path.join(predictions_outdir, f"{sample_idx}.png")
                         )
 
+                processed_samples += len(idx)
+
+                if progress_callback is not None:
+                    progress_callback(processed_samples, total_samples)
+
+                if (
+                    metrics_callback is not None
+                    and evaluation_step is not None
+                    and evaluation_step > 0
+                    and (
+                        processed_samples % evaluation_step == 0
+                        or processed_samples == total_samples
+                    )
+                ):
+                    metrics_callback(
+                        um.get_metrics_dataframe(metrics_factory, eval_ontology),
+                        processed_samples,
+                        total_samples,
+                    )
+
         return um.get_metrics_dataframe(metrics_factory, eval_ontology)
 
     def get_computational_cost(