plotting: some stuff

blearn/blearn.py

@@ -18,7 +18,7 @@
 from src.blender import Blender
 
 def main():
-    b = Blender(config_file="config/config_rhone.yml")
+    b = Blender(config_file="config/config_rural_area_demo.yml")
     b.start()
 
 

blearn/config/config_mountain_area_demo.yml

@@ -15,7 +15,7 @@ output:
   format: JPEG
   render: CYCLES
 paths:
-  number_of_samples: 100
+  number_of_samples: 200
   x_min: 0.0
   x_max: 1400.0
   y_min: 0.0

blearn/config/config_rhone_demo.yml

@@ -16,7 +16,7 @@ output:
   render: CYCLES
   use_second_view: False
 paths:
-  number_of_samples: 100
+  number_of_samples: 200
   x_min: 0.0
   x_max: 200.0
   y_min: 0.0

blearn/config/config_road_corridor_demo.yml

@@ -15,7 +15,7 @@ output:
   format: JPEG
   render: CYCLES
 paths:
-  number_of_samples: 100
+  number_of_samples: 200
   x_min: 0.0
   x_max: 800.0
   y_min: 0.0

blearn/config/config_rural_area_demo.yml

@@ -15,7 +15,7 @@ output:
   format: JPEG
   render: CYCLES
 paths:
-  number_of_samples: 100
+  number_of_samples: 200
   x_min: 0.0
   x_max: 500.0
   y_min: 0.0

radarmeetsvision/interface.py

@@ -232,7 +232,7 @@ def validate_epoch(self, epoch, val_loader, iteration_callback=None):
 
                     # TODO: Expand on this interface
                     if iteration_callback is not None:
-                        iteration_callback(int(sample['index']), depth_prediction)
+                        iteration_callback(sample, depth_prediction)
 
                     if mask is not None:
                         current_results = eval_depth(depth_prediction[mask], depth_target[mask])

scripts/plotting/plot_rgbr.py

@@ -0,0 +1,57 @@
+import argparse
+import matplotlib.pyplot as plt
+import numpy as np
+import cv2
+import radarmeetsvision as rmv
+
+from pathlib import Path
+
+rgbr_dir = None
+
+def prediction_callback(sample, prediction):
+    image = sample['image'].cpu().numpy().squeeze()
+    image = np.moveaxis(image, 0, -1)
+    depth_prior = sample['depth_prior'].cpu().numpy().squeeze()
+
+    image = cv2.resize(image, (640, 480))
+    depth_prior = cv2.resize(depth_prior, (640, 480))
+
+    nonzero_mask = (depth_prior > 0.0)
+    ones = np.ones(nonzero_mask.sum())
+    zeros = np.zeros(nonzero_mask.sum())
+    red = np.vstack((ones, zeros, zeros)).T
+    image[nonzero_mask] = red
+    image = np.clip(image, 0.0, 1.0)
+    plt.imsave(str(rgbr_dir / f"{int(sample['index']):05d}_rgbr.jpg"), image)
+
+def main(args):
+    rmv.setup_global_logger()
+    interface = rmv.Interface()
+    interface.set_encoder('vitb')
+    depth_min = 0.19983673095703125
+    depth_max = 120.49285888671875
+    interface.set_depth_range((depth_min, depth_max))
+    interface.set_output_channels(2)
+
+    interface.set_size(480, 640)
+    interface.set_batch_size(1)
+    interface.set_criterion()
+    interface.set_use_depth_prior(True)
+
+    interface.load_model(pretrained_from=args.network)
+    _, loader = interface.get_single_dataset(args.dataset, min_index=0, max_index=-1)
+
+    global rgbr_dir
+    rgbr_dir = Path(args.dataset) / 'rgbr'
+    if not rgbr_dir.is_dir():
+        rgbr_dir.mkdir(exist_ok=True)
+
+    interface.validate_epoch(0, loader, iteration_callback=prediction_callback)
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description='Purely evalute a network')
+    parser.add_argument('--dataset', type=str, required=True, help='Path to the dataset directory')
+    parser.add_argument('--network', type=str, help='Path to the network file')
+    args = parser.parse_args()
+    main(args)