Improve docs

README.md

@@ -43,6 +43,7 @@ Most of what the scripts do is to wrap input and output to the different modules
     3. Fill-in missing keypoints (as OpenPose sometimes does not output every keypoint) by maintaining the same distance to a connected keypoint (e.g. wirst-elbow) as when the keypoint was last seen.  This increases the accuracy of the classifier later on.
     4. Fill-in missing frames by interpolating positions of every keypoint.  This is done to normalise the data in case OpenPose lost track of a person for a couple of frames.  With this normalisation, every chunk of the same length will also have the same length in video-time.
 * If there are multiple datasets that you wish to combine, you can run the [`combine_datasets.py`](combine_datasets.py) script which allows you to do exactly that.
+* A possible improvement here is to allow a user to label different chunks in the same video with different lengths.
 
 #### visualise_dataset.py
 

action_recognition/analysis/labelling.py

@@ -163,21 +163,20 @@ def pseudo_automatic_labelling(self, timestamps, frames_per_chunk, video, tracks
         Returns
         -------
         chunks : numpy.array
-            The labelled chunks
+            The labelled chunks.
         frames : numpy.array
-            The frames for the labelled chunks
+            The frames for the labelled chunks.
         labels : numpy.array
-            The labels for each chunk
+            The labels for each chunk.
         indicies : numpy.array
             The index of the track for every chunk.
-            needed for reproducibility.
+            Needed for reproducibility.
         """
         keypoints = tracks[0][0].keypoints
         chunk_shape = (frames_per_chunk, *keypoints.shape)
         chunks, frames, labels, track_indicies = self._init_arrays(chunk_shape, frames_per_chunk)
 
         for timestamp in timestamps:
-            # Only include the first track that fits the timestamp
             for i, track in enumerate(tracks):
                 track_arrays = self._pseudo_automatic_labelling(
                     timestamp, track, i, frames_per_chunk, chunk_shape, video)

action_recognition/classifiers/classification_visualiser.py

@@ -21,10 +21,14 @@ def plot_confusion_matrix(self, labels, test_labels, class_names, title):
 
         Parameters
         ----------
-        labels : array-like of predicted labels
-        test_labels : array-like of ground truth labels
-        class_names : array-like of class names.
-        title : str, title of the plot and name of the file
+        labels : array-like
+            Predicted labels from the classifier.
+        test_labels : array-like
+            Ground truth labels.
+        class_names : array-like
+            The names of the classes in the data.
+        title : str
+            Title of the plot and name of the file.
 
         """
         confusion_matrix = metrics.confusion_matrix(test_labels, labels).astype(np.float32)
@@ -47,13 +51,16 @@ def visualise_incorrect_classifications(self, pred_labels, test_labels, chunks,
 
         Parameters
         ----------
-        pred_labels : array-like of predicted labels.
-        test_labels : array-like of ground truth labels.
-        chunks : array-like, the corresponding chunks for the labels.
-        frames : array-like, the frames for the chunks.
-        translated_chuns : array-like
-            the chunks with transforms.TranslateChunks() applied
-        videos : array-like, the paths to the corresponding videos of the chunks.
+        pred_labels : array-like
+            Predicted labels from the classifier.
+        test_labels : array-like
+            Ground truth labels.
+        chunks : array-like
+            The corresponding chunks for the labels.
+        frames : array-like
+            The frame numbers for the chunks.
+        videos : array-like
+            The paths to the corresponding videos of the chunks.
 
         """
         visualiser = ChunkVisualiser(chunks, frames, translated_chunks)

create_dataset.py

@@ -100,7 +100,7 @@ def process_tracks(tracks_file, video, target_frames_per_chunk, overlap_percenta
     np_tracks = tracks_npz['tracks']
     np_frames = tracks_npz['frames']
 
-    logging.info("Combining, cleaning and removing tracks.")
+    logging.info("Combining, cleaning, and removing tracks.")
     processor = PostProcessor()
     processor.create_tracks(np_tracks, np_frames)
     processor.post_process_tracks()