Uncertainty fixes and altered architecture (#171)

* Add query for `evaluate_msssim_sampled`

* Fix `evaluate_msssim_sampled`

* Fix binning for histograms

* Add no_grad architecture

* Towncrier

* Add bins as function argument

* Delete unnecessary conversion to ms ssim

docs/changes/171.bugfix.rst

@@ -0,0 +1,2 @@
+- fix `evaluate_msssim_sampled`
+- fix call of `evaluate_msssim_sampled`

docs/changes/171.maintenance.rst

@@ -0,0 +1 @@
+- set number of bins for histogram plotting

radionets/dl_framework/architectures/res_exp.py

@@ -161,3 +161,66 @@ def forward(self, x):
         x4 = self.elu(x4)
 
         return torch.cat([x0, x3, x1, x4], dim=1)
+
+
+class SRResNet_16_unc_no_grad(nn.Module):
+    def __init__(self):
+        super().__init__()
+
+        self.preBlock = nn.Sequential(
+            nn.Conv2d(2, 64, 9, stride=1, padding=4, groups=2), nn.PReLU()
+        )
+
+        # ResBlock 16
+        self.blocks = nn.Sequential(
+            SRBlock(64, 64),
+            SRBlock(64, 64),
+            SRBlock(64, 64),
+            SRBlock(64, 64),
+            SRBlock(64, 64),
+            SRBlock(64, 64),
+            SRBlock(64, 64),
+            SRBlock(64, 64),
+            SRBlock(64, 64),
+            SRBlock(64, 64),
+            SRBlock(64, 64),
+            SRBlock(64, 64),
+            SRBlock(64, 64),
+            SRBlock(64, 64),
+            SRBlock(64, 64),
+            SRBlock(64, 64),
+        )
+
+        self.postBlock = nn.Sequential(
+            nn.Conv2d(64, 64, 3, stride=1, padding=1, bias=False),
+            nn.InstanceNorm2d(64),
+        )
+
+        self.final = nn.Sequential(nn.Conv2d(64, 4, 9, stride=1, padding=4, groups=2))
+
+        self.hardtanh = nn.Hardtanh(-pi, pi)
+        self.relu = nn.ReLU()
+        self.elu = GeneralRelu(sub=-1e-10)
+
+    def forward(self, x):
+        s = x.shape[-1]
+
+        x = self.preBlock(x)
+
+        x = x + self.postBlock(self.blocks(x))
+
+        x = self.final(x)
+
+        x0 = x[:, 0].reshape(-1, 1, s // 2 + 1, s)
+        # x0 = self.relu(x0)
+        x1 = x[:, 1].reshape(-1, 1, s // 2 + 1, s)
+        # x1 = self.hardtanh(x1)
+        x3 = x[:, 2].reshape(-1, 1, s // 2 + 1, s)
+        with torch.no_grad():
+            x3 = self.elu(x3)
+
+        x4 = x[:, 3].reshape(-1, 1, s // 2 + 1, s)
+        with torch.no_grad():
+            x4 = self.elu(x4)
+
+        return torch.cat([x0, x3, x1, x4], dim=1)

radionets/evaluation/plotting.py

@@ -628,11 +628,10 @@ def plot_box(ax, num_boxes, corners):
         )
 
 
-def histogram_ms_ssim(msssim, out_path, plot_format="png"):
+def histogram_ms_ssim(msssim, out_path, bins=30, plot_format="png"):
     mean = np.mean(msssim)
     std = np.std(msssim, ddof=1)
     fig, (ax1) = plt.subplots(1, figsize=(6, 4))
-    bins = np.arange(msssim.min(), 1 + 0.01, 0.01)
     ax1.hist(
         msssim,
         bins=bins,
@@ -664,12 +663,10 @@ def histogram_ms_ssim(msssim, out_path, plot_format="png"):
     plt.savefig(outpath, bbox_inches="tight", pad_inches=0.01, dpi=150)
 
 
-def histogram_sum_intensity(ratios_sum, out_path, plot_format="png"):
+def histogram_sum_intensity(ratios_sum, out_path, bins=30, plot_format="png"):
     fig, (ax1) = plt.subplots(1, figsize=(6, 4))
     mean = np.mean(ratios_sum)
     std = np.std(ratios_sum, ddof=1)
-    bins = np.arange(0.05, ratios_sum.max() + 0.05, 0.1)
-    bins = np.insert(bins, 0, 0)
     ax1.hist(
         ratios_sum,
         bins=bins,
@@ -703,12 +700,10 @@ def histogram_sum_intensity(ratios_sum, out_path, plot_format="png"):
     plt.savefig(outpath, bbox_inches="tight", pad_inches=0.01, dpi=150)
 
 
-def histogram_peak_intensity(ratios_peak, out_path, plot_format="png"):
+def histogram_peak_intensity(ratios_peak, out_path, bins=30, plot_format="png"):
     fig, (ax1) = plt.subplots(1, figsize=(6, 4))
     mean = np.mean(ratios_peak)
     std = np.std(ratios_peak, ddof=1)
-    bins = np.arange(0.05, ratios_peak.max() + 0.05, 0.1)
-    bins = np.insert(bins, 0, 0)
     ax1.hist(
         ratios_peak,
         bins=bins,
@@ -764,12 +759,10 @@ def histogram_mean_diff(vals, out_path, plot_format="png"):
     plt.savefig(outpath, bbox_inches="tight", pad_inches=0.01, dpi=150)
 
 
-def histogram_area(vals, out_path, plot_format="png"):
+def histogram_area(vals, out_path, bins=30, plot_format="png"):
     vals = vals.numpy()
     mean = np.mean(vals)
     std = np.std(vals, ddof=1)
-    bins = np.arange(0.05, np.round(vals.max()) + 0.05, 0.1)
-    bins = np.insert(bins, 0, 0)
     fig, (ax1) = plt.subplots(1, figsize=(6, 4))
     ax1.hist(
         vals, bins=bins, color="darkorange", linewidth=3, histtype="step", alpha=0.75

radionets/evaluation/scripts/start_evaluation.py

@@ -17,6 +17,7 @@
     evaluate_intensity_sampled,
     evaluate_mean_diff,
     evaluate_ms_ssim,
+    evaluate_ms_ssim_sampled,
     evaluate_point,
     evaluate_unc,
     evaluate_viewing_angle,
@@ -103,7 +104,11 @@ def main(configuration_path):
 
     if eval_conf["ms_ssim"]:
         click.echo("\nStart evaluation of ms ssim.\n")
-        evaluate_ms_ssim(eval_conf)
+        samp_file = check_samp_file(eval_conf)
+        if samp_file:
+            evaluate_ms_ssim_sampled(eval_conf)
+        else:
+            evaluate_ms_ssim(eval_conf)
 
     if eval_conf["intensity"]:
         click.echo("\nStart evaluation of intensity.\n")

radionets/evaluation/train_inspection.py

@@ -564,7 +564,8 @@ def evaluate_ms_ssim_sampled(conf):
     out_path = Path(model_path).parent / "evaluation"
     out_path.mkdir(parents=True, exist_ok=True)
 
-    data_path = str(out_path) + "/sampled_imgs.h5"
+    name_model = Path(model_path).stem
+    data_path = str(out_path) + f"/sampled_imgs_{name_model}.h5"
     loader = create_sampled_databunch(data_path, conf["batch_size"])
     vals = []
 
@@ -586,10 +587,10 @@ def evaluate_ms_ssim_sampled(conf):
     click.echo(f"\nThe mean ms-ssim value is {vals.mean()}.\n")
 
     if conf["save_vals"]:
-        click.echo("\nSaving area ratios.\n")
+        click.echo("\nSaving msssim ratios.\n")
         out = Path(conf["save_path"])
         out.mkdir(parents=True, exist_ok=True)
-        np.savetxt(out / "area_ratios.txt", vals)
+        np.savetxt(out / "msssim_ratios.txt", vals)
 
 
 def evaluate_area_sampled(conf):