radios_shape should account for binning/subsampling

and not rely on the data_reader (too early)

nabu/pipeline/dataset_validator.py

@@ -175,7 +175,7 @@ class DatasetValidatorBase:
         self.binning = (dataset_cfg["binning"], dataset_cfg["binning_z"])
         self.dataset_info._binning = self.binning
         subsampling_factor = dataset_cfg["projections_subsampling"]
-        self.projections_subsampling = subsampling_factor
+        self.dataset_info.subsampling_factor = subsampling_factor
         self.dataset_info._projections_subsampled = self.dataset_info.projections
         self.dataset_info._projs_indices_subsampled = self._projs_indices
         if subsampling_factor > 1:

nabu/pipeline/fullfield/new_pipeline.py

@@ -114,7 +114,12 @@ class NewPipeline:
         # TODO: sanity check (eg. compare to size of radios in dataset_info) ?
         self.delta_z = chunk_shape[1] # includes "margin"
         self.n_angles = self.chunk_shape[0]
-        self.radios_shape = self.chunk_shape # (n_a, n_z, n_x)
+        # (n_a, n_z, n_x)
+        self.radios_shape = (
+            self.chunk_shape[0] // self.dataset_info.subsampling_factor,
+            self.chunk_shape[1] // self.dataset_info.binning[1],
+            self.chunk_shape[2] // self.dataset_info.binning[0],
+        )
 
     def _set_margin(self, margin):
         if margin is None: