pr comment to explain decoder

official/vision/beta/projects/yolo/configs/experiments/darknet53.yaml

@@ -9,6 +9,8 @@ task:
       type: 'darknet'
       darknet:
         model_id: 'darknet53'
+    norm_activation: 
+      activation: 'mish'
   losses:
     l2_weight_decay: 0.0005
     one_hot: True

official/vision/beta/projects/yolo/configs/experiments/darknet53_tfds.yaml

@@ -11,6 +11,8 @@ task:
       type: 'darknet'
       darknet:
         model_id: 'darknet53'
+    norm_activation: 
+      activation: 'mish'
   losses:
     l2_weight_decay: 0.0005
     one_hot: True

official/vision/beta/projects/yolo/dataloaders/classification_input.py

@@ -27,6 +27,9 @@ class Decoder(decoder.Decoder):
       return 
 
   def decode(self, serialized_example):
+    # this is not the best solution, but we encode the image because the parser only works with string encoded jpegs. 
+    # the simple solution is to decode the image to a tensor in the deccoder and to have the parser operate on uint8 
+    # tensors rather than tensor strings, or to have the parser function differently based on the input type. 
     sample_dict = {
                 'image/encoded': tf.io.encode_jpeg(serialized_example['image'], quality=100), 
                 'image/class/label': serialized_example['label'], 

official/vision/beta/projects/yolo/modeling/layers/nn_blocks.py

@@ -6,7 +6,6 @@ import tensorflow.keras.backend as K
 from official.modeling import tf_utils
 
 
-
 @ks.utils.register_keras_serializable(package='yolo')
 class Identity(ks.layers.Layer):
     def __init__(self, **kwargs):