Fixes

README.md

@@ -15,14 +15,15 @@ If you want to make a contribution please follow [the guideline](CONTRIBUTING.md
   ```Shell
   git clone https://github.com/opencv/training_toolbox_caffe.git caffe
   cd caffe
-  git checkout master
+  git checkout develop
   ```
 2. Build the code. Please follow [Caffe instruction](http://caffe.berkeleyvision.org/installation.html) to install all necessary packages and build it.
   ```Shell
+  sudo pip install -r $CAFFE_ROOT/python/requirements.txt
   mkdir build && cd build
   cmake ..
   make -j8
-  # Make sure to include $CAFFE_ROOT/python to your PYTHONPATH.
+  export PYTHONPATH=$PYTHONPATH:$CAFFE_ROOT/python
   # (Optional)
   make runtest -j8
   make pytest

python/action_metrics.py

@@ -535,6 +535,9 @@ def calc_mr_ap(scores, true_positives, false_positives, num_gt, num_images, fppi
 
         return 0.5 * (miss_rates[left_position] + miss_rates[right_position])
 
+    if len(true_positives) == 0 or np.sum(true_positives) == 0:
+        return 1.0, 0.0
+
     sorted_ind = np.argsort(-scores)
     fp_sorted = false_positives[sorted_ind]
     tp_sorted = true_positives[sorted_ind]
@@ -696,7 +699,7 @@ def main():
                                   if len([True for b in bbox_list if b.label == class_id]) > 0]
             class_pred_image_ids = [img_id for img_id, bbox_list in iteritems(predicted_actions)
                                     if len([True for b in bbox_list if b.label == class_id]) > 0]
-            class_num_images = np.sum(np.unique(class_gt_image_ids + class_pred_image_ids))
+            class_num_images = np.sum(len(np.unique(class_gt_image_ids + class_pred_image_ids)))
 
             class_mr, class_ap = calc_mr_ap(class_scores, class_tp, class_fp, class_num_gt_bboxes, class_num_images)
             print('   {}: AP: {:.2f} miss_rate@0.1: {:.2f}'

python/caffe/custom_layers/actions_data_layer.py

@@ -134,6 +134,13 @@ class SampleDataFromDisk(object):
                 total_frames += 1
                 total_objects += len(gt_objects)
 
+        for class_id in self._class_queues:
+            if class_id == ignore_class_id:
+                continue
+
+            if len(self._class_queues[class_id]) == 0:
+                raise Exception('Cannot find frames with {} action label'.format(class_id))
+
         LOG('DataLayer stats: loaded {} frames with {} objects.'.format(total_frames, total_objects))
         self._print_stat(glob_class_counts, self._class_queues, self._ignore_class_id)
 

python/caffe/custom_layers/detections_matcher_layer.py

@@ -220,7 +220,7 @@ class DetMatcherLayer(BaseLayer):
             overlaps = intersection_area / union_area
             overlaps[np.less_equal(union_area, 0.0)] = 0.0
 
-            return 1.0 - intersection_area / union_area
+            return 1.0 - overlaps
 
         matched_detections = {}
         for item_id in gt_data:

python/requirements.txt

@@ -10,7 +10,7 @@ networkx>=1.8.1
 nose>=1.3.0
 pandas>=0.12.0
 python-dateutil>=2.6.0
-protobuf>=2.5.0
+protobuf==2.6.1
 python-gflags>=2.0
 pyyaml>=3.10
 Pillow>=5.1.0