Merge pull request #136 from monocongo/issue_135_filter_darknet_label…

…s_parse_bug

Filter module Darknet labels parse bug fix

README.md

@@ -204,8 +204,11 @@ supports limiting the number of bounding boxes per class type. The filtered data
 will contain annotation files with bounding boxes only for the class labels specified 
 and limited to the number of boxes specified for each class label. For example: 
 ```bash
-$ cvdata_filter --src_annotations /data/darknet --dest_annotations /data/filtered_darknet \
-    --src_images /data/images --dest_images /data/filtered_images \
+$ cvdata_filter --format darknet \
+    --src_annotations /data/darknet \ 
+    --dest_annotations /data/filtered_darknet \
+    --src_images /data/images \
+    --dest_images /data/filtered_images \
     --darknet_labels /data/darknet/labels.txt \
     --boxes_per_class car:6000 truck:6000
 ```

src/cvdata/filter.py

@@ -25,21 +25,34 @@ def _darknet_indices_to_labels(
         darknet_labels_path: str,
 ) -> Dict:
     """
-    TODO
+    Parses a Darknet (YOLO) annotation labels file into a dictionary. The labels
+    file is expected to contain a single class label per line, and the resulting
+    dictionary will contain integer keys beginning at 0, so the first class label
+    will be the value for key 0, the second class label will be the value for key
+    1, etc. For example, the labels file with the following lines:
+
+    dog
+    cat
+    panda
+
+    will result in the following indices to labels dictionary:
+
+    { 0: "dog", 1: "cat", 2: "panda" }
 
     :param darknet_labels_path: path to the file containing labels used in
         the Darknet dataset, should correspond to the labels used in the Darknet
-        annotation file
+        annotation files of the dataset
     :return: dictionary mapping index values to corresponding labels text
     """
 
     index_labels = {}
     with open(darknet_labels_path, "r") as darknet_labels_file:
         index = 0
         for line in darknet_labels_file:
-            darknet_label = line.split()[0]
-            index_labels[index] = darknet_label
-            index += 1
+            if len(line.strip()) > 0:
+                darknet_label = line.split()[0]
+                index_labels[index] = darknet_label
+                index += 1
 
     return index_labels