Fix Issues when running tests on the CI

Several issues relating to running tests on the CI have been fixed.

• Databases are now directly available on the CI. There is no need to load them from the ci-assets package.
• Some tests checking the model transforms were failing due to differences in the environment when testing locally vs on the CI. A statistical approach was used to handle this.

tests/conftest.py

@@ -203,7 +203,10 @@ class DatabaseCheckers:
 
     @staticmethod
     def check_image_quality(
-        datamodule, reference_histogram_file, pearson_coeff_threshold=0.005
+        datamodule,
+        reference_histogram_file,
+        compare_type="equal",
+        pearson_coeff_threshold=0.005,
     ):
         ref_histogram_splits = JSONDatabaseSplit(reference_histogram_file)
 
@@ -226,21 +229,27 @@ class DatabaseCheckers:
                     dataset_sample_index
                 ][0]
 
-                image_tensor = numpy.multiply(image_tensor.numpy(), 255).astype(
-                    int
-                )
-                histogram = numpy.histogram(
-                    image_tensor, bins=256, range=(0, 256)
-                )[0].tolist()
-
-                # We cannot test if histograms are exactly equal because
-                # the torch.resize transform is inconsistent depending on the environment.
-                # assert histogram == ref_hist_data
-
-                # Compute pearson coefficients between histogram and reference
-                # and check the similarity within a certain threshold
-                pearson_coeffs = numpy.corrcoef(histogram, ref_hist_data)
-                assert 1 - pearson_coeff_threshold <= pearson_coeffs[0][1] <= 1
+                histogram = []
+                for color_channel in image_tensor:
+                    color_channel = numpy.multiply(
+                        color_channel.numpy(), 255
+                    ).astype(int)
+                    histogram.extend(
+                        numpy.histogram(
+                            color_channel, bins=256, range=(0, 256)
+                        )[0].tolist()
+                    )
+
+                if compare_type == "statistical":
+                    # Compute pearson coefficients between histogram and reference
+                    # and check the similarity within a certain threshold
+                    pearson_coeffs = numpy.corrcoef(histogram, ref_hist_data)
+                    assert (
+                        1 - pearson_coeff_threshold <= pearson_coeffs[0][1] <= 1
+                    )
+
+                else:
+                    assert histogram == ref_hist_data
 
 
 @pytest.fixture