Merge pull request #233 from DiamondLightSource/autopad_fix

fixing autopadding and tests

Author: dkazanc

httomolibgpu/recon/algorithm.py

@@ -671,4 +671,6 @@ def _take_neg_log_np(data: np.ndarray) -> np.ndarray:
 
 def __estimate_detectorHoriz_padding(detX_size) -> int:
     det_half = detX_size // 2
-    return int(np.sqrt(2 * (det_half**2)) // 2)
+    padded_value_exact = int(np.sqrt(2 * (det_half**2))) - det_half
+    padded_add_margin = int(0.1 * padded_value_exact)
+    return padded_value_exact + padded_add_margin

tests/test_recon/test_algorithm.py

@@ -168,9 +168,9 @@ def test_reconstruct_FBP3d_tomobar_3_detpad_true(
     )
 
     recon_data = recon_data.get()
-    assert_allclose(np.mean(recon_data), -0.00044703347, atol=1e-6)
+    assert_allclose(np.mean(recon_data), -0.00035695144, atol=1e-6)
     assert_allclose(
-        np.mean(recon_data, axis=(0, 2)).sum(), -0.05722028, rtol=1e-06, atol=1e-5
+        np.mean(recon_data, axis=(0, 2)).sum(), -0.045689818, rtol=1e-06, atol=1e-5
     )
     assert recon_data.dtype == np.float32
     assert recon_data.shape == (210, 128, 210)
@@ -257,8 +257,8 @@ def test_reconstruct_CGLS3d_tomobar_detpad_true(
     )
     assert recon_data.flags.c_contiguous
     recon_data = recon_data.get()
-    assert_allclose(np.mean(recon_data), 0.0021116333, rtol=1e-07, atol=1e-6)
-    assert_allclose(np.mean(recon_data, axis=(0, 2)).sum(), 0.27028912, rtol=1e-03)
+    assert_allclose(np.mean(recon_data), 0.0021554278, rtol=1e-07, atol=1e-6)
+    assert_allclose(np.mean(recon_data, axis=(0, 2)).sum(), 0.27589476, rtol=1e-03)
     assert recon_data.dtype == np.float32
 
 
@@ -306,8 +306,8 @@ def test_reconstruct_FISTA3d_tomobar_pd_tv_detpad_true(
     )
     assert recon_data.flags.c_contiguous
     recon_data = cp.asnumpy(recon_data)
-    assert_allclose(np.mean(recon_data), 0.0018328167, rtol=1e-07, atol=1e-6)
-    assert_allclose(np.mean(recon_data, axis=(0, 2)).sum(), 0.23460051, rtol=1e-04)
+    assert_allclose(np.mean(recon_data), 0.0018338546, rtol=1e-07, atol=1e-6)
+    assert_allclose(np.mean(recon_data, axis=(0, 2)).sum(), 0.23473345, rtol=1e-04)
     assert recon_data.dtype == np.float32
 
 

zenodo-tests/test_recon/test_algorithm.py

@@ -125,7 +125,7 @@ def test_reconstruct_FBP3d_tomobar_i12_dataset1_autopad(i12_dataset1: tuple):
     )
     assert recon_data.flags.c_contiguous
     recon_data = recon_data.get()
-    assert_allclose(np.sum(recon_data), 7422.912, rtol=1e-07, atol=1e-6)
+    assert_allclose(np.sum(recon_data), 6669.1274, rtol=1e-07, atol=1e-6)
     assert recon_data.dtype == np.float32
     assert recon_data.shape == (2560, 5, 2560)
 
@@ -154,9 +154,9 @@ def test_reconstruct_LPRec3d_tomobar_i12_dataset1(i12_dataset1: tuple):
     )
     assert recon_data.flags.c_contiguous
     recon_data = recon_data.get()
-    assert isclose(np.sum(recon_data), 9085.442, abs_tol=10**-3)
-    assert pytest.approx(np.max(recon_data), rel=1e-3) == 0.0063419784
-    assert pytest.approx(np.min(recon_data), rel=1e-3) == -0.0062314235
+    assert isclose(np.sum(recon_data), 9628.818, abs_tol=10**-3)
+    assert pytest.approx(np.max(recon_data), rel=1e-3) == 0.006367563270032406
+    assert pytest.approx(np.min(recon_data), rel=1e-3) == -0.0062076798
     assert recon_data.dtype == np.float32
     assert recon_data.shape == (2560, 3, 2560)
 
@@ -185,9 +185,9 @@ def test_reconstruct_LPRec3d_tomobar_i12_dataset1_autopad(i12_dataset1: tuple):
     )
     assert recon_data.flags.c_contiguous
     recon_data = recon_data.get()
-    assert int(np.sum(recon_data)) == 8590
-    assert pytest.approx(np.max(recon_data), rel=1e-3) == 0.006351378
-    assert pytest.approx(np.min(recon_data), rel=1e-3) == -0.00622256
+    assert int(np.sum(recon_data)) == 9147
+    assert pytest.approx(np.max(recon_data), rel=1e-3) == 0.0063818083
+    assert pytest.approx(np.min(recon_data), rel=1e-3) == -0.006200762
     assert recon_data.dtype == np.float32
     assert recon_data.shape == (2560, 3, 2560)
 
@@ -231,7 +231,7 @@ def test_reconstruct_LPRec_tomobar_i13_dataset1(i13_dataset1: tuple):
 
     assert recon_data.flags.c_contiguous
     recon_data = recon_data.get()
-    assert int(np.sum(recon_data)) == 1134
+    assert int(np.sum(recon_data)) == 1149
     assert recon_data.dtype == np.float32
     assert recon_data.shape == (4646, 1, 4646)
 
@@ -300,9 +300,9 @@ def test_reconstruct_LPRec3d_tomobar_i13_dataset2(i13_dataset2: tuple):
     )
     assert recon_data.flags.c_contiguous
     recon_data = recon_data.get()
-    assert int(np.sum(recon_data)) == 3887
-    assert pytest.approx(np.max(recon_data), rel=1e-3) == 0.0105562
-    assert pytest.approx(np.min(recon_data), rel=1e-3) == -0.00835598
+    assert int(np.sum(recon_data)) == 3448
+    assert pytest.approx(np.max(recon_data), rel=1e-3) == 0.010543354
+    assert pytest.approx(np.min(recon_data), rel=1e-3) == -0.008385599
     assert recon_data.dtype == np.float32
     assert recon_data.shape == (2560, 10, 2560)
 
@@ -435,7 +435,7 @@ def test_reconstruct_LPRec3d_tomobar_k11_dataset2(k11_dataset2: tuple):
     )
     assert recon_data.flags.c_contiguous
     recon_data = recon_data.get()
-    assert isclose(np.sum(recon_data), 7689.375, abs_tol=10**-3)
+    assert isclose(np.sum(recon_data), 10865.34, abs_tol=10**-2)
     assert recon_data.dtype == np.float32
     assert recon_data.shape == (2560, 25, 2560)
 
@@ -517,6 +517,6 @@ def test_reconstruct_FISTA3d_tomobar_autopad_k11_dataset2(k11_dataset2: tuple):
     )
     assert recon_data.flags.c_contiguous
     recon_data = recon_data.get()
-    assert isclose(np.sum(recon_data), 1293.3489, abs_tol=10**-3)
+    assert isclose(np.sum(recon_data), 1518.5251, abs_tol=10**-3)
     assert recon_data.dtype == np.float32
     assert recon_data.shape == (2560, 5, 2560)