snapshot...

modules/deeplearning/srcnn_l1b_l2.py

@@ -134,6 +134,13 @@ def upsample(tmp):
     return tmp
 
 
+def upsample_one(tmp):
+    tmp = tmp[slc_y_2, slc_x_2]
+    tmp = resample_2d_linear_one(x_2, y_2, tmp, t, s)
+    tmp = tmp[y_k, x_k]
+    return tmp
+
+
 class SRCNN:
     
     def __init__(self):
@@ -676,11 +683,6 @@ class SRCNN:
 
         pred = self.model([data], training=False)
         self.test_probs = pred
-        pred = pred.numpy()
-        if label_param != 'cloud_probability':
-            pred = denormalize(pred, label_param, mean_std_dct)
-
-        return pred
 
     def run(self, directory, ckpt_dir=None, num_data_samples=50000):
         train_data_files = glob.glob(directory+'data_train_*.npy')
@@ -718,54 +720,40 @@ def run_restore_static(directory, ckpt_dir, out_file=None):
 
 
 def run_evaluate_static(in_file, out_file, ckpt_dir):
-    N_X = N_Y = 10
-
-    slc_x = slice(2, N_X*128 + 4)
-    slc_y = slice(2, N_Y*128 + 4)
-    slc_x_2 = slice(1, N_X*128 + 6, 2)
-    slc_y_2 = slice(1, N_Y*128 + 6, 2)
-    x_2 = np.arange(int((N_X*128)/2) + 3)
-    y_2 = np.arange(int((N_Y*128)/2) + 3)
-    t = np.arange(0, int((N_X*128)/2) + 3, 0.5)
-    s = np.arange(0, int((N_Y*128)/2) + 3, 0.5)
-    x_k = slice(1, N_X*128 + 3)
-    y_k = slice(1, N_Y*128 + 3)
-    x_128 = slice(3, N_X*128 + 3)
-    y_128 = slice(3, N_Y*128 + 3)
 
     sub_y, sub_x = (N_Y * 128) + 10, (N_X * 128) + 10
-    y_0, x_0, = 3232 - int(sub_y/2), 3200 - int(sub_x/2)
+    y_0, x_0, = 3232 - int(sub_y/2), 1100 - int(sub_x/2)
 
     h5f = h5py.File(in_file, 'r')
 
-    grd_a = get_grid_values_all(h5f, 'temp_11_0um_nom')
+    grd_a = get_grid_values_all(h5f, 'super/temp_11_0um')
     grd_a = grd_a[y_0:y_0+sub_y, x_0:x_0+sub_x]
     grd_a = np.where(np.isnan(grd_a), 0, grd_a)
     hr_grd_a = grd_a.copy()
-    grd_a = upsample(grd_a)
-    grd_a = normalize(grd_a, 'temp_11_0um_nom', mean_std_dct)
+    grd_a = upsample_one(grd_a)
+    grd_a = normalize(grd_a, 'super/temp_11_0um', mean_std_dct)
     hr_grd_a = hr_grd_a[y_128, x_128]
 
     # ------------------------------------------------------
-    grd_b = get_grid_values_all(h5f, 'refl_0_65um_nom')
+    grd_b = get_grid_values_all(h5f, 'super/refl_0_65um')
     grd_b = grd_b[y_0:y_0+sub_y, x_0:x_0+sub_x]
     grd_b = np.where(np.isnan(grd_b), 0, grd_b)
     hr_grd_b = grd_b.copy()
     hr_grd_b = hr_grd_b[y_128, x_128]
     # Full res:
-    grd_b = grd_b[:, slc_y, slc_x]
-    grd_b = normalize(grd_b, 'refl_0_65um_nom', mean_std_dct)
+    grd_b = grd_b[slc_y, slc_x]
+    grd_b = normalize(grd_b, 'super/refl_0_65um', mean_std_dct)
 
-    grd_c = get_grid_values_all(h5f, label_param)
+    grd_c = get_grid_values_all(h5f, 'super/'+label_param)
     grd_c = grd_c[y_0:y_0+sub_y, x_0:x_0+sub_x]
     hr_grd_c = grd_c.copy()
     hr_grd_c = np.where(np.isnan(hr_grd_c), 0, grd_c)
     hr_grd_c = hr_grd_c[y_128, x_128]
 
     grd_c = np.where(np.isnan(grd_c), 0, grd_c)
-    grd_c = upsample(grd_c)
+    grd_c = upsample_one(grd_c)
     if label_param != 'cloud_probability':
-        grd_c = normalize(grd_c, label_param, mean_std_dct)
+        grd_c = normalize(grd_c, 'super/'+label_param, mean_std_dct)
 
     data = np.stack([grd_a, grd_b, grd_c], axis=2)
     data = np.expand_dims(data, axis=0)
@@ -773,7 +761,12 @@ def run_evaluate_static(in_file, out_file, ckpt_dir):
     h5f.close()
 
     nn = SRCNN()
-    out_sr = nn.run_evaluate(data, ckpt_dir)
+    nn.run_evaluate(data, ckpt_dir)
+    out_sr = nn.test_probs
+    out_sr = out_sr.numpy()
+    if label_param != 'cloud_probability':
+        out_sr = denormalize(out_sr, label_param, mean_std_dct)
+
     if out_file is not None:
         np.save(out_file, (out_sr[0, :, :, 0], hr_grd_a, hr_grd_b, hr_grd_c))
     else: